Download Lemons, Market Shutdowns and Learning

Lemons, Market Shutdowns and Learning
Pablo Kurlat
∗
Stanford University
October 2010
Abstract
I study a dynamic economy featuring adverse selection in financial markets. Investment
is undertaken by borrowing-constrained entrepreneurs. They sell their past projects to
finance new ones, but asymmetric information about project quality creates a lemons
problem. The magnitude of this friction responds to aggregate shocks, amplifying the
responses of asset prices and investment. Indeed, negative shocks can lead to a complete
shutdown in financial markets. I then introduce learning from past transactions. This
makes the degree of informational asymmetry endogenous and makes the liquidity of
assets depend on the experience of market participants. Market downturns lead to less
learning, worsening the future adverse selection problem. As a result, transitory shocks
can create highly persistent responses in investment and output. (JEL E22, E44, D83,
G14)
∗
I am deeply indebted to George-Marios Angeletos, Ricardo Caballero, Bengt Holmström and especially Iván
Werning for their invaluable guidance. I also thank Daron Acemoglu, Sergi Basco, Francisco Buera, Martı́n
Gonzalez-Eiras, Guido Lorenzoni, Julio Rotemberg, Jean Tirole, Robert Townsend and participants at the MIT
Macroeconomics Lunch, Theory Lunch and International Breakfast, as well a various seminar participants.
Correspondence: Department of Economics, Stanford University. 579 Serra Mall, Stanford, CA, 94305. Email:
[email protected]
1
1
Introduction
Financial markets are fragile, volatile and occasionally shut down entirely. The recent financial
crisis has intensified economists’ interest in understanding the causes of financial instability
and its effects on real economic variables such as investment, output and productivity. In this
paper I develop a model of financial imperfections to explain how instability in general and
market shutdowns in particular can result from macroeconomic shocks and in turn amplify and
propagate them.
I focus on one specific type of financial market imperfection: asymmetric information about
the quality of assets. This particular imperfection is worth studying for several reasons. First,
the ability of creditors to seize a debtor’s assets, either as a possible equilibrium outcome or
as an off-equilibrium threat, is crucial for enabling financial transactions to take place, both
in theory (Hart and Moore 1994, Kiyotaki and Moore 1997) and in everyday practice. If
there is asymmetric information about asset qualities, which is a natural assumption, this has
the potential to interfere with a large subset of financial transactions. Second, asymmetric
information is a central concern in corporate finance. Following Myers and Majluf (1984),
asymmetry of information between firm managers and their outside investors is seen as a key
determinant of firms’ capital structure. Third, sometimes financial markets simply cease to
function, as documented for instance by Gorton and Metrick (2009) for the repo market in
2007-2009. Since Akerlof (1970), it is well known that the complete breakdown of trade is a
theoretical possibility in economies with asymmetric information. This means that asymmetric
information at least has the potential to explain extreme crises and may shed light on less
extreme phenomena as well.
I embed imperfect financial markets in a simple dynamic macroeconomic model. In the
model, entrepreneurs hold the economy’s stock of capital. Every period, they receive random
idiosyncratic investment opportunities. The only way to obtain financing is by borrowing
against existing assets or, equivalently, selling them. Assets are bought by entrepreneurs who
in the current period have poor investment opportunities but nevertheless wish to save part of
their dividends. Unfortunately, some fraction of existing assets are useless lemons and buyers
can’t tell them apart from high quality assets (nonlemons), creating a classic lemons problem.
I show that the lemons problem introduces a wedge between the return on saving and the
cost of funding, persuading some entrepreneurs to stay out of the market. This is formally
equivalent to introducing a tax on financial transactions, with revenues rebated lump-sum to
entrepreneurs. This defines a notion of liquidity where the degree of illiquidity of assets is the
size of the implicit tax. The tax lowers asset prices, the rate of return obtained by uninformed
investors and the rate of capital accumulation. Furthermore, the implicit tax depends on
the proportions of lemons and nonlemons sold, which respond to aggregate shocks. Standard
2
productivity shocks increase current dividends, which increases the supply of savings and raises
asset prices. This persuades more entrepreneurs to sell their nonlemons, improving the overall
mix of projects that get sold and lowering the implicit tax on financial transactions. Shocks to
the productivity of investment have similar effects because they increase entrepreneurs’ desire
to invest and thus their willingness to sell nonlemons. The endogenous response of the size of
the tax implies that asymmetric information can be a source of amplification of the effects of
shocks on both capital accumulation and asset prices. Large negative shocks may lead financial
markets to shut down entirely.
The model predicts that capital becomes more liquid in economic expansions. This prediction is consistent with empirical research by Eisfeldt and Rampini (2006), who find that the
costs of reallocating capital across firms are countercyclical. It is also consistent with the evidence in Choe et al. (1993), who find that the negative price reaction to an offering of seasoned
equity is smaller and the number of firms issuing equity is larger in the expansionary phase of
the business cycle, suggestive of countercyclical adverse selection costs.
In reality, asymmetric information does not mean that relatively uninformed parties do
not know anything. Instead, it can be a matter of degree. In order to investigate how the
degree of asymmetry could vary endogenously, I extend the baseline model by introducing
public information about the quality of individual assets. Each asset issues a signal which is
correlated with its true quality. The correlation is imperfectly known and changes over time.
The precision of entrepreneurs’ estimates of the correlation determines how informative they
find the signals. They learn about the current value of the correlation by observing samples
of past transactions. More transactions lead to larger sample sizes, more precise estimates,
more informativeness of future signals and lower future informational asymmetry. Conversely,
market shutdowns lead to smaller sample sizes, less certainty about the correlation between
signals and quality and more severe informational asymmetry in the future. This can be a
powerful propagation channel by which temporary negative shocks can lead to financial crises
followed by long recessions, in a manner consistent with the evidence in Cecchetti et al. (2009),
Claessens et al. (2009) and Cerra and Saxena (2008).
The learning mechanism in the model formalizes the notion that assets will be more liquid if
they are more familiar and familiarity depends on experience. Accumulated financial experience
is a form of intangible social capital which increases liquidity and reduces frictions in the
investment process. Learning-by-doing in financial markets plays the important role of building
up that capital.
Because investment opportunities are heterogeneous, the distribution of physical investment
across entrepreneurs matters for capital accumulation. Asymmetric information lowers the level
of investment of entrepreneurs with relatively good opportunities, who may decide not to sell
their existing nonlemons due to depressed prices or receive lower prices if they do sell them.
3
At the same time, it increases the level of investment of entrepreneurs with relatively poor
investment opportunities, who might decide to undertake them anyway rather than buy assets
from others for fear of receiving lemons. These effects lower the average rate of transformation
of consumption goods into capital goods and thus can be seen as a determinants of endogenous
investment-sector-specific productivity. Shocks to this type of productivity have been found to
be an important driver of output fluctuations in estimated quantitative models by Greenwood
et al. (2000), Fisher (2006) and Justiniano et al. (2008a). Furthermore, Justiniano et al.
(2008b) find that movements in investment-sector productivity are correlated with measures of
the smooth functioning of financial markets, as would be predicted by the model.
Measurement of investment sector productivity depends on accurate measures of capital
formation. If these fail to take into account changes in the the efficiency of investment due
to changes in the degree of informational asymmetry, information effects in one period would
show up as measured Solow residuals in future periods. Thus the movement of implicit tax
wedges in the model can be a source of changes in (measured) TFP or, in the terminology of
Chari et al. (2007), movements in efficiency wedges.
In common with Kiyotaki and Moore (1997), Bernanke and Gertler (1989), Bernanke,
Gertler and Gilchrist (1999) and Carlstrom and Fuerst (1997) among others, financial frictions in my model are sensitive to wealth effects. However, what governs their severity in my
model is not the wealth of financially-constrained investors, since the margin that determines
whether to sell or keep nonlemons is independent of wealth. Instead, the wealth of those who
finance them matters because it governs the demand for assets.
The structure of the model is close to that developed by Kiyotaki and Moore (2005, 2008),
which also features random arrival of investment opportunities, borrowing constraints and
partially illiquid assets. Those papers use a reduced-form model of the limitations on selling
capital and investigate whether this may explain why easier-to-sell assets command a premium.
In contrast, I develop an explicit model of what the sources of these limitations are, which allows
me to investigate how they respond to aggregate shocks. My model also shares some of the
simplifying assumptions of the Kiyotaki-Moore framework, in particular that entrepreneurs
have no labour income and log preferences. One additional simplification that I make is that
physical capital is the only asset. Thanks to this assumption, entrepreneurs’ policy functions
can be found in closed form despite having an infinite dimensional state vector (due to the
continuum of possible signals) and a nonlinear budget set. This makes it possible to derive
most of the qualitative results analytically and to simulate the model at little computational
cost.
Following (Stiglitz and Weiss 1981), adverse selection played an important early role in the
theory of credit markets, although the emphasis was on the riskiness of projects rather than the
quality of assets. Financial market imperfections that arise specifically due to a lemons problem
4
in asset quality have recently been studied by Bolton et al. (2009) and Malherbe (2009). These
papers model games where dynamic strategic complementarities can give rise to different types
of equilibria, with more or less severe adverse selection. Instead in my model the equilibrium
is unique so the severity of the lemons problem responds to aggregate shocks in a predictable
way.
The lemons problem in macroeconomic settings has been studied by Mankiw (1986), de Meza
and Webb (1987) and House (2006) among others. Closest to this paper is Eisfeldt (2004). In
her model, entrepreneurs hold different vintages of projects and cannot diversify risks. The
reason financial transactions are desirable is that they enable entrepreneurs to smooth consumption when they suffer poor realizations of income from previous vintages of risky projects.
Thus her paper is about how asymmetric information interferes with risk-sharing whereas mine
is about how it interferes with the financing of investment. On a more technical side, one
limitation of her approach is that it requires keeping track of of the distribution of portfolio
holdings across different vintages of projects, for all entrepreneurs, which makes it necessary to
limit attention to numerical simulations of steady states or simple deterministic cycles, since
stochastic simulations are computationally infeasible.
The idea that economic recessions are associated with reduced learning is explored by Veldkamp (2005), van Nieuwerburgh and Veldkamp (2006) and Ordoñez (2009). In these models,
what agents need to learn about is the state of aggregate productivity. The speed of learning
governs how fast output and prices align themselves with fundamentals but the direction of
this alignment is just as likely to be towards higher or lower output. In contrast, in my model
agents learn about parameters of the information structure. More learning alleviates informational asymmetries, which helps the functioning of financial markets for any given level of
productivity. Another difference is that in my model the activity which generates information
is selling projects rather undertaking them. Since the volume of financial transactions can be
very volatile, this opens the door to strong learning effects.
The rest of the paper is organized as follows. Section 2 introduces the model and section
3 and describes frictionless benchmarks. Section 4 describes the equilibrium conditions under
asymmetric information and contains the results for the model without signals. Section 5
contains the extension of the model with signals and learning. Section 6 offers some brief final
remarks. Proofs are collected in Appendix B.
2
The environment
Households. There are two kinds of agents in the economy, workers and entrepreneurs. There
is a continuum of mass L of identical workers, each of whom supplies one unit of labour
inelastically; they have no access to financial markets, so they just consume their wage. In
5
addition, there is a continuum of mass one of entrepreneurs, indexed by j, who have preferences
E
∞
X
β t u(cjt )
t=0
with u(cjt ) = log(cjt ). They do not work.
Technology. Final output (coconuts) is produced combining capital and labour. The capital
stock consists of projects owned by entrepreneurs. Entrepreneur j’s holdings of projects are
R
denoted ktj so the aggregate capital stock is Kt = ktj dj. Every period a fraction λ of projects
becomes useless or “lemons”. Each entrepreneur’s holdings of projects is sufficiently well diversified that the proportion λ applies at the level of the individual entrepreneur as well. Each
of the (1 − λ) Kt projects that do not become lemons is used for production, so that output
is Yt = Y ((1 − λ) Kt , L; Zt ), where Y is a constant-returns-to-scale production function that
satisfies Inada conditions and Zt is exogenous productivity. The marginal product of capital
and labour are denoted YK and YL respectively.
The aggregate resource constraint is
Lcw
t
+
Z
cjt + ijt dj ≤ Y ((1 − λ) Kt , L; Zt )
(1)
j
j
where cw
t denotes consumption per worker, ct is consumption by entrepreneur j and it represents
physical investment by entrepreneur j.
Physical investment is undertaken in order to convert coconuts into projects for period
t + 1. Each entrepreneur can transform coconuts into projects using an idiosyncratic linear
technology with a stochastic marginal rate of transformation Ajt . In addition, each nonlemon
project turns into γ projects at t + 1, so it is possible to interpret 1 − γ (1 − λ) as an average
rate of depreciation. Aggregate capital accumulation is given by
Kt+1 = γ (1 − λ) Kt +
Z
ijt Ajt dj
(2)
Ajt is iid across entrepreneurs and across periods and is drawn from a distribution F with finite
mean.
Allocations. The exogenous state of the economy is zt ≡ Zt , Āt . It includes productivity
Zt and the function Āt , which maps each entrepreneur to a realization of Ajt . An allocation
specifies consumption and investment for each agent in the economy and aggregate capital after
every history: {cw (z t ) , cj (z t ) , ij (z t ) , K (z t )}. An allocation is feasible if it satisfies constraints
(1) and (2) for every history given some K0 .
Information. At time t each entrepreneur knows which of his own projects have become
6
lemons in the current period, but the rest of the agents in the economy do not. In section
5, I augment the model by allowing for publicly observable signals about individual projects.
Informational asymmetry lasts only one period. At t + 1, everyone is able to identify the
projects that became lemons at t, so they effectively disappear from the economy, as illustrated
in figure 1. This assumption is made for simplicity as it eliminates the need to keep track
of projects of different vintages. Daley and Green (2009) study the strategic issues that arise
when informational asymmetries dissipate gradually over time.
Period t
Period t + 1
Lemon
λ
project disappears
1 project
λ
1−λ
Nonlemon
enters Yt (·)
γ projects
Lemon
1−λ
Nonlemon
enters Yt+1 (·)
Figure 1: Information about a project over time
The investment opportunity Ajt is and remains private information to entrepreneur j.
3
Symmetric information benchmarks
3.1
Complete Arrow-Debreu markets
Suppose zt and the quality of individual projects were public information and there were complete competitive markets. The price of lemons will be zero so it is enough to focus on factor
markets and trades of coconuts for nonlemons and state-contingent claims.
Factor markets are competitive. Entrepreneurs hire workers at a wage of w (z t ) = YL (z t )
coconuts and obtain dividends of r (z t ) = YK (z t ) coconuts for each nonlemon project.1 Coconuts are traded for nonlemon projects, ex-dividend, at a spot price of pN L (z t ) coconuts per
nonlemon project. State-contingent claims are traded one period ahead: the state-price density
for obtaining a coconut in history (z t , zt+1 ) is ρ(z t , zt+1 ).
An entrepreneur who starts with k0j projects solves the following program:
1
As is standard, this could be the result of competitive firms renting capital from entrepreneurs or of
entrepreneurs operating the productive technology themselves. With asymmetric information, the latter interpretation avoids the need to analyze adverse selection in the rental market.
7
max
t
{c(z t ),k(z t ),dNL (z ),b(z t ,zt+1 ),i(z t )}
s.t.
c z
t
+ pN L (z t )dN L z
t=0
β tu c zt
+ E ρ z t , zt+1 b z t , zt+1
≤ r z t (1 − λ) k z t + b z t−1 , zt
k z t , zt+1 = γ (1 − λ) k z t + dN L z t + Aj (zt ) i z t
i z t ≥ 0, dN L (z t ) ≥ −(1 − λ)k(z t )
#
"
t−1
Y
ρ(z s , zs+1 ) ≥ 0
lim E b(z t )
+i z
t
E
∞
X
t→∞
t
(3)
(4)
(5)
(6)
(7)
s=0
Constraint (4) is the entrepreneur’s budget constraint in terms of coconuts. The entrepreneur’s available coconuts are equal to the dividends from his nonlemons r (z t ) (1 − λ)
k (z t ) plus net state-contingent coconuts bought the previous period b (z t−1 , zt ). These are
used for consumption plus physical investment plus net purchases of nonlemons dN L (z t ) plus
purchases of state-contingent coconuts for period t + 1. Constraint (5) keeps track of the entrepreneur’s holdings of projects. k (z t , zt+1 ), the total number of projects the entrepreneur
has in history (z t , zt+1 ), is equal to the nonlemon projects he owned at the end of period z t ,
which were (1 − λ) k (z t ) + dN L (z t ), and have grown at rate γ plus the projects that result from
his physical investment in the previous period, Aj (zt ) i (z t ). Constraint (6) states that investment must be nonnegative and sales of nonlemons are limited by the number of nonlemons the
entrepreneur owns. Constraint (7) is a no-Ponzi condition.
The first order conditions with respect to i (z t ) and dN L (z t ) imply
Aj (zt ) ≤
γ
, with equality if i z t > 0
t
pN L (z )
(8)
Let Amax be the highest possible value of A. By the law of large numbers, at each history
there will be an entrepreneur (the best entrepreneur ) with Aj = Amax who can transform each
γ
coconut into Amax projects at t + 1.2 Equation (8) then implies that pN L (z t ) = Amax
for all z t .
At each history, the best entrepreneur is the only one to undertake physical investment. He
finances this investment by selling claims to coconuts one period ahead (i.e. borrowing) which
he then satisfies with the dividends plus proceeds of selling the newly created projects in the
spot market. Since r (z t ) is stochastic, capital is a risky asset, and the best entrepreneur will
2
With a continuous F there will be a zero measure of best entrepreneurs, but a positive measure of entrepreneurs with Aj ∈ (Amax − δ, Amax ] for any positive δ. The results below follow from taking the limit as
δ → 0.
8
use state-contingent securities to share this risk with the rest of the entrepreneurs. Complete
markets imply that risk-sharing will be perfect.
Complete markets are actually not indispensable for achieving the complete markets allocation. All that is needed is that there exist a market for selling new projects at the same
instant that they are built, before anyone knows whether they will become a lemon the follow1
, the best
ing period. In equilibrium these new projects will trade at a price pN EW (z t ) = Amax
entrepreneur will be the only one to invest and entrepreneurs will bear no idiosyncratic risk.
Their only asset in any given period will consist of projects, so they will automatically share
aggregate risk in proportion to their wealth. Since they have identical homothetic preferences,
this coincides with what they would do with complete markets.
Proposition 1. If there are complete markets, all the physical investment is undertaken by the
entrepreneur with Aj = Amax ; all entrepreneurs obtain a return of Amax projects per coconut
saved and they bear no idiosyncratic risk. The same allocation is obtained if the only market
that exists is for newly-created projects.
The aggregate economy behaves just like an economy where the rate of transformation of
coconuts into projects is fixed at Amax , there is a representative entrepreneur and workers are
constrained to live hand-to-mouth. Due to log preferences, it is straightforward to compute the
entrepreneur’s consumption choice, which will be given by
j
c z
t
h
= (1 − β) (1 − λ) YK z t +
γ i
Amax
k j (z t )
(9)
and hence aggregate capital accumulation will be:3
K Z t , Zt+1 = β (1 − λ) Amax YK Z t + γ K Z t
3.2
(10)
Borrowing constraints with symmetric information
For various reasons, it may be difficult for an entrepreneur to borrow against his future wealth,
i.e. to choose negative values of b (z t , zt+1 ). For instance, he may be able to run away with
his wealth rather than honouring his debts.4 Creditors’ main means of enforcing their claims
is the threat to seize the entrepreneur’s assets. In other words, the entrepreneur’s assets serve
as collateral for any obligations he undertakes. Kiyotaki and Moore (2008) point out that it
is important to distinguish between assets that are already in place at the time the financial
transaction is initiated and those that are not, since the latter are harder for creditors to keep
3
Equations (9) and (10) assume that the nonnegativity of aggregate investment is not binding. Otherwise
c (z t ) = YK (z t )k j (z t ) and K (Z t , Zt+1 ) = γ(1 − λ)K(Z t ).
4
Alternatively, he could refuse to exert effort if he has pledged the output to someone else, as in Holmström
and Tirole (1998).
j
9
track of and subject to more severe moral hazard problems. In what follows I make the extreme
assumption that entrepreneurs can costlessly run away with coconuts and hide new projects
from their creditors, which makes them useless as collateral. However, they cannot hide projects
that already exist a the time the transaction is initiated. These constitute the only form of
collateral.
Collateralized financial transactions could take many forms. However, in this model the
future payoffs of current nonlemon projects are binary: either they become a lemon at t + 1
or they do not. Any financing transaction must therefore have zero repayment if the project
becomes a lemon and positive repayment otherwise. If there is no aggregate risk, this makes
selling the asset and using it as collateral for borrowing exactly equivalent.5 Selling is simpler to model, so I assume that the only kind of transaction is ex-dividend sales of existing
projects. This is intended to represent the wider range of transactions that use existing assets
as collateral.6
The entrepreneur will solve program (3), with the added constraint:
b z t , zt+1 ≥ 0
(11)
Constraint (11) will bind for the best entrepreneur. As a result, he will not be able to
undertake all the investment in the economy. Instead, there will be a cutoff A∗ (z t ) = pNLγ(z t )
such that entrepreneurs with Aj (z t ) < A∗ (z t ) will not invest and entrepreneurs with Aj (z t ) >
A∗ (z t ) will sell all their existing nonlemons in order to obtain coconuts for investment.
This equilibrium is inefficient in two related ways. First, the economy does not exclusively
use the most efficient technology (Amax ) for converting coconuts into projects. The best entrepreneur is financially constrained and thus unable to invest all the coconuts the economy
saves, so others with Aj ∈ (A∗ (z t ) , Amax ) also undertake physical investment. Secondly, entrepreneurs are exposed to idiosyncratic risk. If they draw a low value of Aj , they must convert
their coconuts into projects through the market, which only provides a return A∗ (z t ), whereas
if they draw a higher value they convert them at a rate Aj (z t ).
4
Asymmetric information
Assume, as in section 3.2, that the only transactions in financial markets are sales of existing
projects. However, now there is asymmetric information: only the owner of a project knows
whether it is a lemon, and each entrepreneur observes only his own Aj . Those who purchase
5
If there is aggregate risk, selling the asset is equivalent to state-contingent borrowing proportional to the
value that the asset would have in each state of the world if it does not become a lemon.
6
In Kurlat (2009) I study the case of a general joint distribution of asset qualities and investment opportunities and allow for arbitrary contracts.
10
projects have rational expectations about λM (z t ), the proportion of lemons among the projects
that are actually sold in the market.7
With asymmetric information, entrepreneurs do not observe the state of the economy zt =
Zt , Āt because they do not observe the mapping Āt . Instead, they observe an individual
state ztj ≡ Zt , Ajt which includes aggregate productivity Zt and their own idiosyncratic
investment-productivity draws. Fortunately, the endogenous aggregate variables r, p and λM
that are relevant for the entrepreneur’s decisions depend only on the history of productivity
Z t , which is part of the entrepreneur’s information set and hence the entrepreneur’s problem
can be specified in terms of the history of individual states z j,t . An entrepreneur who starts
with k0j projects solves the following program:
max
j
),i(z j,t ),sL (z j,t ),sNL (z j,t ),d(z j,t )}
{c(z j,t ),k(z j,t ,zt+1
E
∞
X
t=0
β t u c z j,t
(12)
s.t.
c z
+i z
+ p Z t d z j,t − sL z j,t − sN L z j,t ≤ r Z t (1 − λ) k z j,t
j k z j,t , zt+1
= γ (1 − λ) k z j,t + 1 − λM Z t d z j,t − sN L z j,t + Aj ztj i z j,t
i z j,t ≥ 0, sL z j,t ∈ 0, λk z j,t , sN L z j,t ∈ 0, (1 − λ) k z j,t , d z j,t ≥ 0
j,t
j,t
Program (12) incorporates the borrowing constraint (11) and the fact that the price p (Z t )
applies equally for sales of lemons sL (z j,t ), sales of nonlemons sN L (z j,t ) and purchases of
projects of unknown quality d (z j,t ), a proportion λM (Z t ) of which turn out to be lemons.
I will look for a recursive competitive equilibrium with X ≡ {Z, Γ} as a state variable,
where Γ(kt , At ) is the cumulative distribution of entrepreneurs over holdings of capital and investment opportunities.8 The relevant state variable for entrepreneur j’s problem is {k j , Aj , X}
so (dropping the j superscript) he solves the following recursive version of program (12):
7
One might still ask why an entrepreneur cannot sell claims against his entire portfolio of projects (by the
law of large numbers, he is not asymmetrically informed about it) instead of selling them individually. Kiyotaki
and Moore (2003) assume that it is possible to credibly bundle all of one’s projects by paying some cost. I
assume this cost is prohibitively large.
8
Since Aj is iid, then it is independent of k j and Γ is just the product of F and the distribution of k. The
more general formulation could easily accommodate the case where an entrepreneur’s individual Aj has some
persistence, which would create some correlation between k j and Aj .
11
V (k, A, X) =
max
c,k ′ ,i,sL ,sNL ,d
[u (c) + βE [V (k ′ , A′ , X ′) |X]]
(13)
s.t.
c + i + p (X) [d − sL − sN L ] ≤ r (X) (1 − λ) k
k ′ = γ (1 − λ) k + 1 − λM (X) d − sN L + Ai
i ≥ 0, d ≥ 0
sL ∈ [0, λk] , sN L ∈ [0, (1 − λ) k]
Denote the solution to this program by {c (k, A, X) , k ′ (k, A, X) , i (k, A, X) , sL (k, A, X) ,
sN L (k, A, X) , d (k, A, X)} and define the supply of lemons and nonlemons, total supply of
projects and demand of projects respectively as
SL (X) ≡
SN L (X) ≡
Z
Z
sL (k, A, X) dΓ(k, A)
sN L (k, A, X) dΓ(k, A)
S (X) ≡ SL (X) + SN L (X)
Z
D (X) ≡
d (k, A, X) dΓ(k, A)
Definition 1. A recursive competitive equilibrium consists of prices {p (X) , r (X) , w (X)};
market proportions of lemons λM (X); a law of motion Γ′ (X) and associated transition density
Π (X ′ |X); a value function V (k, A, X) and decision rules {cw (X) , c (k, A, X) , k ′ (k, A, X) ,
i (k, A, X) , sL (k, A, X) , sN L (k, A, X) , d (k, A, X)} such that (i) factor prices equal marginal
products: w (X) = YL (X), r (X) = YK (X); (ii) workers consume their wage cw (X) =
w (X); (iii) {c (k, A, X), k ′ (k, A, X), i (k, A, X), sL (k, A, X), sN L (k, A, X), d (k, A, X)} and
V (k, A, X) solve program (13) taking p (X), r (X), λM (X) and Π (X ′ |X) as given; (iv) the
market for projects clears: S (X) ≥ D (X), with equality whenever p (X) > 0; (v) the marL (X)
ket proportion of lemons is consistent with individual selling decisions: λM (X) = SS(X)
and
R
(vi) the law of motion of Γ is consistent with individual decisions: Γ′ (k, A)(X) = k′(k̃,Ã,X)≤k
dΓ(k̃, Ã) F (A)
4.1
Solution of the entrepreneur’s problem and equilibrium conditions
I solve the entrepreneur’s problem and find equilibrium conditions in steps. First I show that
all the policy functions are linear in k, which implies an aggregation result. Second I show
12
that, given choice of c and k ′ , the choices of d, sL , sN L and i reduce to a simple arbitrage
condition. Third I solve a relaxed problem, converting the entrepreneur’s nonlinear budget set
into a weakly larger linear one and show that there is a simple static characterization of the
consumption-savings decision. Based on the solution to the relaxed problem it is possible to derive supply, demand and a market clearing condition. Finally I show that the equilibrium price
must satisfy the market-clearing condition whether or not the solutions to the two programs
coincide. In either case the rest of the equilibrium objects follow immediately.
Linearity of policy functions. The constraint set in program (13) is linear in k and the
utility function is homothetic. Hence the policy functions c (k, A, X), k ′ (k, A, X), i (k, A, X),
sL (k, A, X), sN L (k, A, X) and d (k, A, X) are all linear in k. This implies the following aggregation result:
Lemma 1. Prices and aggregate quantities do not depend on the distribution of capital holdings,
only on total capital K.
By Lemma 1, {Z, K} is a sufficient state variable; in order to compute aggregate quantities
and prices it is not necessary to know the distribution Γ.
Buying, selling and investing decisions. Take the choice of k ′ as given. The entrepreneur’s
problem then reduces to choosing d, sL , sN L and i to maximize c. This program is linear so
the entrepreneur will generically choose corner solutions. The decision to keep or sell lemons
is trivial: as long as p > 0 the entrepreneur will sell the lemons (sL = λk), since they are
worthless to him if kept. The decisions to keep or sell nonlemons and to invest in new projects
or in purchasing projects depend on A. The return (i.e. the number of t + 1 projects obtained
γ (1−λM )
per coconut spent) from buying projects is AM ≡
. I refer to this as the market rate
p
of return.9 Conversely, the number of t + 1 nonlemon projects an entrepreneur must give up
to obtain one coconut is γp > AM . The return on investing is simply A. This implies that the
optimal choices of d, sN L and i are given by two cutoffs, shown in figure 2.
AM ≡
Buyer:
Keep nonlemons
Buy projects
γ(1−λM )
p
γ
p
Keeper:
Keep nonlemons
Invest
A
Seller:
Sell nonlemons
Invest
Figure 2: Buying, selling and investing decision as a function of A
Suppose first that k ′ > γ (1 − λ) k so the entrepreneur wants to save more than by just
keeping his nonlemons. For A < AM , entrepreneurs are Buyers: the return from buying is
9
Noting, however, that it involves two different goods (projects and coconuts) as well as two different dates.
13
greater than the return from investing so i ≥ 0 and sN L ≥ 0 bind and d > 0. For A ∈ [AM , γp ]
entrepreneurs are Keepers: investing offers a higher return than buying but not higher than
the opportunity cost of selling nonlemons at the market price, so the entrepreneur neither buys
projects nor sells nonlemons; d ≥ 0 and sN L ≥ 0 bind and i > 0. For A > γp entrepreneurs are
Sellers: the return from investing is high enough for the entrepreneurs to sell nonlemons in order
to finance investment; d ≥ 0 and sN L ≤ (1 − λ) k bind and i > 0. If instead k ′ < γ (1 − λ) k
(which by lemma 4 below is inconsistent with equilibrium), then Buyers and Keepers would
choose i = d = 0 and sN L > 0 while Sellers would choose d = 0, sN L = (1 − λ) k and i > 0.
Combining these arbitrage conditions with the constraint from program (13) yields the following
lemma:
Lemma 2. Given k ′ , the optimal d, sL , sN L and i are given by
i
Buyers: A ∈ 0, AM
Keepers: A ∈ AM , γp
sL =
λk
d=
max
sN L =
i=
λk
max
0
n
n
k ′ −γ(1−λ)k
γ(1−λM )
,0
γ(1−λ)k−k ′
,0
γ
o
o
0
max
max
Sellers: A ∈
λk
0
n
n
γ(1−λ)k−k ′
,0
γ
k ′ −γ(1−λ)k
A
,0
o
o
γ
,∞
p
(14)
(1 − λ) k
k′
A
Consumption-savings decision under a relaxed budget set. An entrepreneur with investment
′
opportunity is A must choose kc and kk from his budget set, shown in figure 3.
k′
k
A[(1 − λ)r + p]
Seller
(1 − λ) γ+
[(1 − λ) r + λp] A
Keeper
(1 − λ) γ+
[(1 − λ) r + λp] AM
Buyer
(1 − λ)γ
x
(1 − λ)r
+λp
(1 − λ)r
+p
c
k
Figure 3: Budget sets
Point x represents an entrepreneur who chooses sL = λk and i = sN L = d = 0, an option
14
available to all entrepreneurs. He simply consumes the dividends (1 − λ) rk and the proceeds
from selling lemons λpk, and enters period t + 1 with (1 − λ) γk projects.
Consider first the decision of a Keeper. If he wishes to increase consumption beyond point
x he must sell nonlemons, which means giving up γp projects for each additional coconut of
consumption. If instead he wishes to carry more projects into t+1, he invests with productivity
A. Hence the budget constraint is kinked: to the right of x the slope is − γp whereas to the left
it is −A. Consider next a Buyer. His budget set is the same as for the Keeper except that the
return he obtains from saving beyond point x is the market return AM , which is higher than
his individual return on investment A but lower than that of Keepers. Lastly, a Seller will sell
all his projects and his budget constraint is linear with constant slope −A.
Define the entrepreneur’s virtual wealth as
W (k, A, X) ≡ λp (X) + (1 − λ) r (X) + max p (X) ,
γ
max {A, AM (X)}
k
(15)
Virtual wealth corresponds to to the projection of the left half of the budget constraint onto
the horizontal axis. It consists of the coconuts the entrepreneur has (dividends plus proceeds of
selling lemons) plus the nonlemon projects, valued at the maximum of either their sale
h price p or i
′
γ
their replacement cost max{A,AM (X)} . The linear budget set kk ≤ max A, AM (X) W (k,A,X)
− kc
k
is weakly larger than the true kinked budget, so substituting it in program (13) leads to the
following relaxed program:
V (k, A, X) = max
[u (c) + βE [V (k ′ , A′ , X ′) |X]]
′
c,k
(16)
s.t.
k ′ = max A, AM (X) [W (k, A, X) − c]
Lemma 3. Under the relaxed program (16), the entrepreneur’s consumption is c (k, A, X) =
(1 − β) W (k, A, X)
Due to logarithmic preferences, entrepreneurs will always choose to consume a fraction
1 − β of their virtual wealth and save the remaining β, by some combination of keeping their
old nonlemons, buying projects and physical investment. Note that the entrepreneur’s decision,
while rational and forward looking, does not depend on the transition density Π (X ′ |X) or on
the stochastic process for A. This feature will make it possible to solve for the equilibrium
statically.
Supply and demand under the relaxed program. Take p as given. By (14), the supply of
15
projects will include all the lemons plus the nonlemons from Sellers. Hence
γ
S (p) = λ + (1 − λ) 1 − F
K
p
(17)
This implies a market proportion of lemons of
λM (p) =
λ
λ + (1 − λ) 1 − F γp
(18)
and a market rate of return of:10
γ
(1
−
λ)
1
−
F
γ
p
γ
M
M
A (p) =
1 − λ (p) =
p
p λ + (1 − λ) 1 − F γ
p
(19)
Demand for projects will come from Buyers. By Lemma 3, under the relaxed program they
′ −γ(1−λ)k
choose k ′ =βAM W k, AM , X . By Lemma 2, they each demand k γ(1−λ
projects. Using
M)
(15) and adding over all Buyers, demand for projects will be:
r
(1 − β) (1 − λ)
D (p) = β λ + (1 − λ)
−
F AM (p) K
M
p
1 − λ (p)
(20)
Market clearing implies
S(p∗ ) ≥ D(p∗ ) with equality whenever p∗ > 0
(21)
Equilibrium conditions under the true program.
Lemma 4. D > 0 only if the solutions to programs (13) and (16) coincide for all entrepreneurs
The solutions to programs (13) and (16) will not coincide whenever in the relaxed program,
some entrepreneurs wish to choose points to the right of x. Lemma 4 states that if this is the
case there will be no demand for projects and the price must be zero.
This implies the following result:
Proposition 2.
1. In any recursive equilibrium, the function p (X) satisfies (21) for all X
2. For any p(X) that satisfies (21), there exists a recursive competitive equilibrium where
the price is given by p(X)
10
Define A(0) ≡ 0.
16
3. There exists at least one function p(X) that satisfies (21)
Proposition 2 establishes that a recursive equilibrium exists and must satisfy (21) regardless
of whether or not the solutions to programs (13) and (16) coincide. Therefore it is possible to
find equilibrium prices statically simply by solving (21). Once p∗ is determined, it is straightforward to solve, also statically, for the rest of the equilibrium objects. λM and AM follow
from (18) and (19). If p∗ > 0 then virtual wealth and, by Lemma 3, consumption for each
entrepreneur can be found using (15) and sL , sN L , d and i are given by (14). If instead the
only solution to (21) is p∗ = 0, I refer to the situation as one of market shutdown. It is still
possible to solve the relaxed problem (16), which results in
k ′ = β (1 − λ) (Ar + γ) k
This satisfies k ′ ≥ γ (1 − λ) k iff A ≥ Ā ≡ γr (1−β)
. Hence for entrepreneurs with A ≥ Ā,
β
consumption and investment can be computed in the same way as when the market does not
shut down whereas entrepreneurs with A < Ā chose c = (1 − λ) rk and k ′ = γ (1 − λ) k.
Aggregate capital accumulation is found by replacing the equilibrium values of i into the
law of motion of capital (2),11 yielding
Z γ
p
K′
[βA [λp + (1 − λ) r] − (1 − β) (1 − λ) γ] dF (A)
= γ (1 − λ) +
K
AM
Z ∞
βA [p + (1 − λ) r] dF (A)
+
(22)
γ
p
In general, the market return AM (p) can be either increasing or decreasing in p. An increase
in the price has a direct effect of lowering returns by making projects more expensive and an
indirect effect of improving returns by increasing the proportion of entrepreneurs who choose to
sell their nonlemons. This implies that there could be more than one solution to (21). In this
case, I will assume that the equilibrium price is given by the highest solution. More worryingly,
there could exist a price p′ > p∗ such that AM (p′ ) > AM (p∗ ) even when p∗ is the highest solution
to (21). This will be the case when selection effects are strong enough that the return from
buying projects would be higher at a price higher than the highest market-clearing one. Both
Buyers and Sellers would be better off if there was sufficient demand to sustain such a price.
Stiglitz and Weiss (1981) argue that when this is the case the equilibrium concept used above
is not reasonable and it would be more sensible to assume that Buyers set a price above p∗ that
maximizes their return and ration the excess supply. Appendix A discusses how the definition
of equilibrium may be adapted to allow for rationing, a change that makes little difference
for the results. In section 5, I consider signals that segment the market into a continuum of
11
By Walras’ Law, it is equivalent to just sum k ′ (k, A, X) over all entrepreneurs.
17
different submarkets. In that variant there exists a unique equilibrium in which Buyers can
never benefit from raising prices in any submarket (see Lemma 7), so the issue of what is the
right equilibrium concept becomes moot, a point first made by Riley (1987) in the context of
the Stiglitz-Weiss model.12 For some of the results below, it will simplify the analysis to simply
assume that parameters are such that the issue does not arise:
Assumption 1. AM (p) is decreasing
A sufficient condition for Assumption 1 to hold is h (x) ≤
1
x
1+
1−λ
λ
(1 − F (x)) for all x,
where h is the hazard function of A. Results do not rely on Assumption 1 unless otherwise
stated.
4.2
Liquidity/Risk Premia
Entrepreneurs in the model do not face a portfolio problem. If an entrepreneur wishes to
carry wealth from one period to the next, the only way to do it is to buy or build projects.
M
For each coconuthe saves, he obtains
max
A,
A
projects at t + 1. By (15), Wk (A, X) =
n
o
γ
λp (X) + (1 − λ) r (X) + max p (X) , max{A,AM (X)}
is the marginal value of a project, so
saving one coconuts yields the equivalent of a (risky) amount of max A, AM Wk (A′ , X ′ )
coconuts at t + 1. It is possible to define the implicit risk-free rate Rf for a given entrepreneur
by assuming he has access to an alternative safe technology that converts t-dated coconuts into
t + 1-dated coconuts (and hence faces a portfolio problem) and asking what the return on that
technology would need to be for him not to change his equilibrium decisions.
Formally, consider an entrepreneur who has access to a technology that delivers R coconuts
tomorrow in exchange for a coconut today. The entrepreneur solves:13
V (W, A, X) = max′ [u (c) + βE [V (W ′ , A′ , X ′ ) |X]]
c,π,W
s.t.
W ′ = π max A, AM Wk (A′ , X ′ ) + (1 − π) R (W − c)
π ∈ [0, 1]
where π is the fraction of his savings W − c that he invests in projects. Define Rf (A, X) as the
maximum value of R such that π = 1 is optimal given {A, X}.
Proposition 3.
12
In their terminology, there will be redlining (exclusion of arbitrarily similar yet distinct groups) but not
pure rationing (partial exclusion of observationally identical projects).
13
This is the relaxed program. Proposition 3 below easy generalizes to the case where this does not necesarily
coincide with the full program.
18
1. Rf (A, X) < max A, AM E [Wk (A′ , X ′ )] for all {A, X}
2. Under symmetric information and deterministic X ′ , then Rf = max A, AM E [Wk (A′ , X ′ )]
Proposition 3 states that for any entrepreneur the implicit risk free rate is lower than the
expected return (measured in coconuts) of investing in projects. This premium arises because
Wk is decreasing in A. The higher an entrepreneur’s investment productivity, the lower the
marginal value he places on an existing project. Furthermore, Lemma 3 implies that agents
who value projects the least also consume less, so project valuation is negatively correlated
with the marginal utility of consumption. The premium would disappear under symmetric
information as agents trade away their different relative valuations for projects.
Kiyotaki and Moore (2008) obtain a similar result by assuming that resaleability constraints
prevent entrepreneurs from reselling a fraction of their projects. Here instead the difference
between the values placed on projects by entrepreneurs with different investment opportunities
is derived endogenously as a result of asymmetric information.
4.3
Equivalence with an economy with taxes
As shown in figure 2, asymmetric information introduces a wedge between the return obtained
by Buyers, AM , and the return given up by Sellers of nonlemons, γp . The magnitude of this
wedge depends on λM (X). It turns out that this wedge is exactly isomorphic to the wedge
that would be introduced by imposing state-dependent taxes on the sales of projects.
Consider the economy with borrowing constraints and symmetric information of section 3.2,
but now assume that the government imposes an ad-valorem tax of τ (X) p coconuts on sales
of projects. The total revenue T (X) = τ (X)p(X)S(p(X)) collected from this tax is rebated
to entrepreneurs in proportion to their capital holdings. Entrepreneurs solve the following
program:
V (k, A, X) =
max
c,k ′ ,i,sNL ,d
[u (c) + βE [V (k ′ , A′ , X ′ ) |X]]
(23)
s.t.
c + i + p (X) [d (1 + τ (X)) − sN L ] ≤ r (X) (1 − λ) k + T (X)
k ′ = γ [(1 − λ) k + d − sN L ] + Ai
i ≥ 0, d ≥ 0
sN L ∈ [0, (1 − λ) k]
This problem can be solved by the same steps used to solve program (13). Solving for the
equilibrium conditions leads to the following equivalence result.
19
Proposition 4. Suppose τ (X) =
λM ∗ (X)
,
1−λM ∗ (X)
where λM ∗ (X) is the equilibrium value of the
asymmetric information economy. Then prices and allocations of the symmetric-informationwith-taxes and the asymmetric information economies are identical.
By Proposition 4, the distortionary effect of having a proportion λM of lemons in the market
M
λ
is exactly equivalent to the one that would result from a tax at the rate τ = 1−λ
M . Moreover,
asymmetric information gives all entrepreneurs the possibility of earning λpk coconuts from
selling lemons to others. This has an exact counterpart in the pro-rata redistribution of the
government’s revenue.
Chari et al. (2007) propose a way to decompose economic fluctuations into the movements
of an efficiency wedge, a labour wedge, an intertemporal wedge and a government spending
wedge. The implicit taxes that result from asymmetric information do not translate neatly into
a single one of these wedges. They distort both the consumption-saving decision (resulting in an
intertemporal wedge) and the allocation of investment across different entrepreneurs (resulting
in an efficiency wedge). Furthermore, the model would have an intertemporal wedge even
under symmetric information due to borrowing constraints and the fact that workers do not
participate in asset markets.
It is reasonably simple to analyze the effects of exogenous changes in tax rates. This will
be useful when looking at the economy with asymmetric information because Proposition 4
implies these are exactly isomorphic to the effects of endogenous changes in λM .
Lemma 5. For any state X
1.
dp
dτ
2.
dAM
dτ
3.
dK ′ dτ τ =0
<0
<0
<0
An increase in taxes increases the wedge between AM and γp . Parts 1 and 2 of Lemma 5 establish that this increase in the wedge manifests itself through both lower returns for Buyers and
lower prices for Sellers. Both of these effects tend to lower capital accumulation. In addition,
taxes have the effect of redistributing resources from Buyers and Sellers to all entrepreneurs,
including Keepers. As with any tax, the relative incidence on Buyers and Sellers depends on
elasticities. For small enough τ , the elasticities of supply and demand are mechanically
linked,
M
as the density of marginal Buyers, f A , approaches that of marginal Sellers, f γp . Part
3 of Lemma 5 establishes that in this case the redistributive effect always goes against the
higher-A agents, reinforcing the effect of lower capital accumulation.14
14
For τ away from zero, it is possible to construct counterexamples where f
20
γ
p
is much higher than f AM ,
4.4
Comparative statics and aggregate shocks
The equilibrium conditions derived in section 4.1 are static. This feature is a consequence of
assuming that entrepreneurs have log preferences, no labour income and a single asset to invest
in. This simplifies the analysis of the effects of aggregate shocks. In particular, it implies that
shocks will have the same effects whether or not they are anticipated. Thus by answering the
comparative statics question “how would the features of the model change if a parameter were
different?” one also answers the impulse response question “how would the economy respond
to a shock to one of the parameters?”
Consider first the effects of a productivity shock. It makes a difference in this model whether
the shock affects primarily the coconut-producing capacity of the economy or its projectproducing capacity. Suppose first that there is a proportional shock to coconut-productivity.
This would affect the equilibrium conditions through its effect on the marginal product of capital r = YK . Its effects can therefore be understood through comparative statics with respect
to r.
Proposition 5. If in equilibrium p∗ > 0 then
1. p∗ is increasing in r.
2. Under Assumption 1, AM ∗ is decreasing in r.
3. λM ∗ is decreasing in r.
4. Under Assumption 1,
K′
K
is increasing in r
Favourable shocks will mean that entrepreneurs receive higher current dividends and hence
hold a larger number of coconuts. Other things being equal, entrepreneurs would want to save
a fraction β of the additional coconuts. Sellers and Keepers would do so through physical
investment but Buyers would attempt to buy more projects, thus bidding up the price (part 1)
and lowering returns (part 2). Note that it is not productivity per se that matters but rather the
effect of the productivity shock on current dividends. A similar effect would result, for instance,
if there was a temporary shock to the capital share of output leaving total output unchanged or
simply a helicopter drop of coconuts from outside the economy. Part 3 of Proposition 5 has the
important implication that the severity of the lemons problem, as measured by the equivalent
M
λ
tax wedge τ = 1−λ
M will respond to aggregate shocks. Higher prices persuade marginal Keepers
to sell their nonlemons and therefore a favourable shock to the coconut-producing capacity of
the economy will alleviate the lemons problem.
so supply is much more elastic than demand. In this case it is possible for Sellers to be net beneficiaries of
redistribution, so taxes can conceivably increase capital accumulation. Working in the opposite direction is the
fact that the direct marginal distortion increases with τ .
21
Turn now to an investment-productivity shock. This can be represented as a proportional
change in the investment opportunity of every
entrepreneur, from A to φA, so that the distri-
bution of A becomes F ′ , where F ′ (A) = F
A
φ
.
Proposition 6.
1. λM ∗ is decreasing in φ.
2. Under Assumption 1,
K′
K
is increasing in φ
Proposition 6 implies that higher productivity in the project-producing sector also alleviates
the lemons problem. In this case, the effect comes from the supply side rather than the demand
side. Because physical investment has become more attractive, marginal Keepers decide to sell
their nonlemons, improving the mix of projects.
Propositions 5 and 6 jointly show that positive shocks lessen financial market wedges and
negative shocks worsen them. Liquidity, as measured by (the inverse of) the size of these
wedges, is procyclical.
For negative shocks, the adverse selection effect can be sufficiently strong to lead to a
complete shutdown of financial markets.
Proposition 7.
1. If
max AM (p) <
p
γ (1 − β)
r
β
(24)
then the market shuts down
2. Sufficiently large negative shocks to coconut-productivity or project-productivity lead to
market shutdowns
When r is low, entrepreneurs have very few coconuts for each project they own. This raises
the return that is needed to entice Buyers to choose k ′ above the kink in figure 3. When
project-productivity is low, the measure of entrepreneurs who are willing to sell nonlemons
at any given price becomes low. λM increases, lowering returns. In either case, if there was
symmetric information, in equilibrium asset prices would drop low enough to increase returns
to the point where Buyers have sufficient demand to clear the market. With asymmetric
information, however, the adverse selection effect places an upper bound on the return AM (p).
If either kind of productivity is low enough, the return that would be needed to have positive
demand for projects is higher than this upper bound and the market shuts down.
It is also possible to analyze shocks whose only effect is due to informational asymmetry.
Consider a temporary increase in λ, compensated by an increase in K such that (1 − λ) K
22
remains unchanged. This shock has no effect on the production possibility frontier of the
economy and, with symmetric information, would have no effect on allocations. With asymmetric information, the fact that there are more lemons mixed in with the nonlemons makes a
difference.
Proposition 8. An simultaneous increase in λ and K that leaves (1 − λ) K unchanged increases λM ∗
The increase in λM ∗ that results from this type of shock is equivalent to an increase in
taxes, so the results in Lemma 5 regarding the effects on asset prices, rates of return and
capital accumulation can be applied directly.
One interpretation of this type shock may be the following. Suppose every period entrepreneurs receive an endowment of ∆λ K useless lemons, so the total number of lemons
is (λ + ∆λ ) K rather than λK. However, in ordinary times it is possible to tell apart the
endowment-lemons from the nonlemons, so their existence is irrelevant. A shock to λ of the
kind described above is equivalent to entrepreneurs losing the ability to detect endowmentlemons, a form of deterioration of information. Effects of this sort will play a role in section 5
where I make the quality of information endogenous.
The endogenous response of liquidity has the important consequence of amplifying the response of the economy to exogenous shocks. To show this, I compare the responses of economies
with symmetric and asymmetric information to the same exogenous shock. In order to make
sure that the economies are otherwise identical, I assume that in the symmetric information
economy there are (fixed) taxes on transactions at a rate such that, absent the shock, prices
and allocations in both economies would be exactly the same. Denote equilibrium variables in
both economies by the superscripts SI and AI respectively.
Proposition 9.
1.
dpAI
dr
2.
dAM,SI
dr
3.
dK ′,AI
dr
>
dpSI
dr
<
>
dAM,AI
dr
dK ′,SI
dr
for λ small enough
Proposition 9 implies that, in response to a productivity shock which increases r, asymmetric information amplifies the rise in asset prices, moderates the drop in rates of return and
amplifies the increase in the rate of capital accumulation compared to the symmetric information benchmark.
The idea that asset prices may affect productive but financially constrained agents and that
this can amplify shocks is of course not new. What is newer to this model is that, rather than
23
being a fixed parameter as in Kiyotaki and Moore (2005, 2008), the degree of illiquidity of assets
itself responds to shocks, which is a related but slightly different channel and is consistent with
the evidence in Eisfeldt and Rampini (2006).
One related prediction of the model is that, because in recessions financial markets work
less well, firms (in the model, entrepreneurs) will rely less on them. Thus in recessions they
will finance a higher proportion of their investment with retained earnings and in expansions
they will rely more on external financing (in the model, selling old projects) in expansions.
This is in contrast to some other models of financial frictions. For instance, in Bernanke and
Gertler (1989), the key to the “financial accelerator” mechanism is that in good times firms
have abundant retained earnings and therefore need to rely less on outside financing.
4.5
Simulations
In this section I compute dynamic examples of the response of the economy to shocks. In order
to highlight the role of asymmetric information, I compare the impulse responses to those of an
economy with a fixed level of taxes on financial transactions such that steady state allocations
are identical.
While I choose parameter values that are close to those used in quantitative models, the
spirit of the exercise is to illustrate the mechanisms underlying the results stated above and
give a rough idea of the potential magnitudes rather than to constitute a quantitatively precise
simulation.
Parameter
β
γ
λ
F (A)
Y
L
Z
Value
0.94
.95
0.05
Gamma distribution with E (A) = 1 and std (A) = 1.75
Z [(1 − λ) K]α L1−α with α = 0.4
1
1
Table 1: Parameter values used in simulations
The production function is a standard Cobb-Douglas. The length of the period is approximately one year. The value of λ is low to make sures that markets do not shut down in steady
state. The values of β and γ lead to a steady-state investment-to-GDP ratio of 0.15 and a
depreciation rate of 0.1. Under these parameters, Assumption 1 holds in the relevant range
where equilibria take place.
In all cases I assume the economy begins at a steady state and is hit by a shock at t = 3.
The first exercise is to simulate a productivity shock in the consumption goods sector lasting
24
a single period. Panel 2 of figure 4 shows the response of output to the TFP shock. It rises
1. TFP
2. Y
1
0.5
2
4
6
0.6
1.5
0.4
1
0.5
0
8
t
4. p
4
6
−0.2
8
1
0
2
4
2
6
8
6
8
6
8
0.5
0
−2
−4
4
t
M
6. A
2
% dev from ss
% dev from ss
2
t
5. Implicit tax rate
2
−1
0.2
0
% dev from ss
0
3. Net K growth
2
%
Asym info
Fixed tax
1.5
% dev from ss
% dev from ss
2
2
4
t
6
t
8
0
−0.5
−1
−1.5
2
4
t
Figure 4: Transitory productivity shock in the consumption goods sector
mechanically at the time of the shock and remains slightly above steady state because of capital
accumulation. Panel 3 shows how capital accumulation responds to the shock, illustrating the
amplifying effect of asymmetric information. Panel 4 illustrates the response of asset prices.
Because the increase in the marginal product of capital increases the supply of savings, these
would rise even with symmetric information; they rise even more because of the selection effect.
The response of the implicit tax rate is shown on panel 5. Panel 6 shows the response of market
rates of return. The increase in the supply of savings drives them down, but selection effects
moderate the effect. After the shock is over at t = 4, the marginal product of capital is slightly
below its steady state level due to diminishing marginal product, so the effects are reversed.
If the productivity shock followed an AR(1) process (with persistence of 0.9), the effects
would be similar to the nonpersistent shock. The main difference is that output in the asymmetric information economy remains above that of the fixed-tax economy for longer due to the
sum of several periods of more capital accumulation.
The next exercise is to simulate a productivity shock in the investment sector, i.e. a shift
in F (A) of the kind considered in Proposition 6, lasting only one period. The most interesting
difference compared to the standard TFP shock lies in the response of AM and p. The increase
in investment-productivity means that more entrepreneurs wish to sell their assets to obtain
financing. Even with symmetric information, this raises market returns, as shown on panel 6
of figure 6. The selection effect means that this is even stronger with asymmetric information.
With symmetric information, the increase in the supply of assets necessarily lowers asset prices,
25
1. TFP
2. Y
1
Asym info
Fixed tax
0.5
0
2
4
6
0.4
1.5
0.2
%
1.5
1
0
0.5
0
8
t
4. p
2
4
6
−0.2
8
0
2
4
6
0
−2
−4
8
6
8
6
8
0.5
% dev from ss
% dev from ss
1
4
t
M
6. A
2
−1
2
t
5. Implicit tax rate
2
% dev from ss
3. Net K growth
2
% dev from ss
% dev from ss
2
2
4
6
t
0
−0.5
−1
−1.5
8
2
4
t
t
Figure 5: AR(1) productivity shock in the consumption goods sector
1. E(A)
2. Y
1
0.5
2
4
6
0.6
0.15
0.4
0.1
0.05
0
8
t
4. p
2
4
6
8
−0.2
0
−0.5
2
4
2
6
8
6
8
6
8
2
0
−2
−4
4
t
6. AM
2
% dev from ss
% dev from ss
0
t
5. Implicit tax rate
0.5
−1
0.2
% dev from ss
0
3. Net K growth
0.2
%
Asym info
Fixed tax
1.5
% dev from ss
% dev from ss
2
2
4
t
6
t
8
1
0
−1
2
4
t
Figure 6: Transitory productivity shock in the investment sector
26
as shown on panel 4. With asymmetric information the increase in the proportion of nonlemons
moderates the effect, and for some parameter values could even reverse it.
5
Informative signals and learning
In section 4 I made the extreme assumption that potential buyers do not know anything about
an individual project. In reality there are many sources of information about assets that
potential buyers may consult, such as financial statements and analyst reports. All of these
are imperfect and the degree to which a prospective buyer finds them informative may depend
on his expertise with that particular kind of information. In this section I extend the model
by introducing both imperfect signals of project quality and the learning process by which
expertise in interpreting them is developed. I then examine how the learning process makes
shocks propagate through the economy.
The structure of information is as shown in figure 7.
Index
Outcome
Signals
µl
l, Blue
Lemon
λ
l
1−λ
Nonlemon
1 − µl
l, Green
µl
l, Green
1 − µl
l, Blue
Figure 7: Information structure
Each project receives a random index l uniformly distributed in [0, 1]. After it has either
become a lemon or not it emits a publicly observable message s ∈ {Blue, Green}.15 A vector µ
governs the conditional probability of emitting each of the two messages for a given index l; µl ≡
Pr [s = Blue|l, Lemon] = Pr [s = Green|l, Nonlemon]. The index l, as well as the messages
Blue or Green are publicly observable, so a signal consists of a pair l, s ∈ [0, 1]×{Blue, Green}.
The amount of information that a signal l, s conveys about project quality depends on µl .
If µl = 12 , then the signals are completely uninformative, whereas if µl = 1 or µl = 0, then
15
To abstract from issues of strategic release of information, I assume that both the index l and the message
s are beyond the entrepreneur’s control.
27
they perfectly communicate the project’s quality. In general, the quality of information (as
measured for instance by the mutual information) is a monotonic function of |µl − 12 |.
Signals have the effect of segmenting the market into a continuum of separate submarkets,
with prices pl,s and proportions of lemons λM
l,s . Denote the price vector by p = {pl,s }l=[0,1],s=B,G
and the lemons-proportion vector by λM = λM
l,s l=[0,1],s=B,G . Entrepreneurs might decide to
sell their nonlemons when they fall into certain l, s submarkets but not others16 and may also
decide which submarkets to buy from.
The index l represents the different pieces of information that a firm can issue in any given
period: its financial statements, news about a labour dispute, consumer reports about its
products, etc. The message s represents the actual content of that piece of information, such as
“the product ranks third in the industry in customer satisfaction” or “market share increased
from 17% to 22% in the past year but profit margins declined”. Potential buyers may use the
signals to infer the probability that a given project is a lemon.
In reality, inferring the the true value of a project on the basis of such information is a
difficult task. A firm that increases market share at the expense of profit margins could be
healthy (a nonlemon) if building market share is valuable because customers in that market
are loyal or it could be struggling (a lemon) if market share increased due to low prices while
costs are increasing. If investors are inexperienced they will find it difficult to assess which
of these explanations is more likely and will therefore find the signal relatively uninformative.
Instead, experienced investors will have observed firms in this industry increase market share
at the expense of profits several times in the past and will know how frequently this turned
out successfully. They may have to worry, however, about whether their experience continues
to be relevant or whether changes in the environment have rendered it obsolete.
I model this tension by assuming that the vector µ, which governs the correct interpretation
of the signals, changes over time and at any point in time entrepreneurs do not know its true
value. Formally, assume that for each l, µl,t follows an independent two-state Markov process,
taking values µ̄ > 12 and 1 − µ̄, with switching probability σ < 12 . Entrepreneurs do not
directly observe the value of µt , but must instead infer it from observing past projects. After
each period, they observe a sample of size Nl of signal-outcome pairs for each index l. Each
observation consists of the signal the project issued plus whether it turned out to be a lemon
or not. Denote the entire set of samples by the random variable χ.
By observing χ, after every period entrepreneurs will form updated beliefs about the true
value of µ. Since all entrepreneurs observe χ, beliefs will be common to all of them. Beliefs
about each µl,t are given by distributions Bl,t in [0, 1] with mean µ̂l,t . Given that the Markov
16
Assume they are sufficiently well diversified that, at the level of the individual entrepreneur, their holdings
of projects are uniformly distributed across l and the proportion of messages Blue and Green for each l is given
by µl .
28
processes for µl,t are independent, beliefs will be such that the Bl,t distributions are independent.
Denote the the overall beliefs by the joint distribution Bt . The size of the sample they observe,
Nl , will determine how precise these beliefs are.
I assume Nl is random and follows a Poisson distribution with mean
ωl = [fl ωS + (1 − fl )ωK ]
(25)
where fl is the fraction of l-indexed projects which have been sold in the period and ωS and
ωK are parameters, with ωS > ωK . Equation (25) says that, for each project, there is a Poisson
probability that the market finds out what happened to it. This probability is higher for
projects that were sold than for projects that were kept by their owner.
The rationale for the assumption that ωS > ωK is that firms that raise funds from the
market usually provide investors with much more detailed information about their financial
condition than those that do not, both at the time of raising funds and thereafter. Part of
this is due to legal reasons, such as reporting requirements for publicly held companies, and
part may be because firms are purposefully attempting to alleviate the lemons problem. The
information that investors observe after investing gives them feedback about how accurate their
assessment of the firm was at the time they decided whether to invest in it. Furthermore, it
is not sufficient that information exist, someone must take the trouble to analyze it order to
learn from it. The main reason someone would do that is to help them decide whether to trade.
When the volume of trade decreases, the amount of attention paid to analyzing information is
likely to decrease as well. Anecdotal evidence certainly suggests that this is the case. To take
just one example, the investment bank Paribas laid off its entire Malaysian research team in
1998 in response to reduced business during the Asian crisis.17
As is standard in rational expectations equilibria, I assume that in addition to χ, agents are
able to observe prices and, if these are informative about the function µ, they can simultaneously
update their beliefs and adjust their demand accordingly.18 They are not, however, able to
observe the quantity of projects traded in any given submarket. Furthermore, they do not
learn µ from observing the signals emitted by the lemons and nonlemons in their own portfolio
of projects.19
In order to formulate the entrepreneur’s problem, I expand the state variable in the economy
to include beliefs B, so X = {Z, Γ, B}. Taking the process for X as given, an entrepreneur
17
Wall Street Journal, October 26, 1998.
It will turn out that prices are not informative about µ, see Lemma 8 below.
19
This may seem inconsistent with the fact that they are fully diversified. However, full diversification can
be achieved by holding a countably infinite number of projects. Learning the true µl for a countable number of
indices l would provide information about a zero-measure subset of submarkets.
18
29
solves:
V (k, A, X) =
max
c,k ′ (µ),i,sL,l,s (µ),sNL,l,s (µ),dl,s
[u (c) + βE [V (k ′ (µ), A′ , X ′) |X, p(X, µ)]]
(26)
s.t.
c+i+
Z
1
0
X
pl,s (X, µ) [dl,s − (sL,l,s (µ) + sN L,l,s (µ))] dl ≤ (1 − λ) r (X) k
s
"
k ′ (µ) = γ (1 − λ) k +
Z 1X
0
#
1 − λM
l,s (X, µ) dl,s − sN L,l,s dl + Ai
s
i ≥ 0, dl,s ≥ 0
sL,l,B ∈ [0, λµl k] , sL,l,G ∈ [0, λ (1 − µl ) k]
sN L,l,B ∈ [0, (1 − λ) (1 − µl ) k] , sN L,l,G ∈ [0, (1 − λ) µl k]
Notice that the entrepreneur does not know λM because how many lemons end up in each
submarket depends on the true µl,s , which is unknown to the entrepreneur. In case dl,s is
positive for some submarket l, s, then λM
l,s determines how many nonlemons the entrepreneur
obtains from that purchase. Therefore the value of k ′ that the entrepreneur will achieve depends
on µ and the entrepreneur may have uncertainty about it. Expectations about the future,
including future beliefs and k ′ are formed knowing the current state X and any information
the current prices provide about current µ.
Definition 2. A recursive equilibrium with signals consists of prices {p (X, µ) , r (X) , w (X)};
market proportions of lemons λM (X, µ); laws of motion Γ′ (X, µ) and B ′ (X, µ, χ)) and associated transition density Π (X ′ |X); a value function V (k, A, X) and decision rules {cw (X) ,
c (k, A, X) , k ′ (k, A, X;µ), i (k, A, X) , sL,l,s (k, A, X; µ) , sN L,l,s (k, A,X;µ), ds (k, A, X; µ)} such
that (i) factor prices equal marginal products: w (X) = YL (X), r (X) = YK (X); (ii) workers
consume their wage cw (X) = w (X); (iii) {c (k, A, X) , k ′ (k, A, X; µ), i (k, A, X) , sL,l,s (k, A, X; µ) ,
sN L,l,s (k, A, X; µ) , dl,s (k, A, X; µ)} and V (k, A,X) solve program (26) taking p (X; µ), r (X),
λM (X; µ) and Π (X ′ |X) as given; (iv) each submarket l, s clears: Sl,s (X; µ) ≥ Dl,s (X; µ),
with equality whenever pl,s (X; µ) > 0; (v) in each market the proportion of lemons is consistent
SL,l,s (X;µ)
with individual selling decisions: λM
l,s (X; µ) = Sl,s (X;µ) ; (vi) the law of motion of Γ is consisR
tent with individual decisions: Γ′ (k, A)(X) = k′ (k̃,Ã,X)≤k dΓ(k̃, Ã) F (A) and (vii) beliefs evolve
according to Bayes’ rule.
Notice one subtlety about the definition of equilibrium. Consistent with the assumption
that Buyers do not know µ, program (26) does not allow the choice of dl,s to depend on
the realization of µ. However, in general program (26) may have many solutions if Buyers
are indifferent between buying from different submarkets (making demand a correspondence
30
rather than a function). If demand is indeed a correspondence, the definition of equilibrium
allows D(X; µ) to take any value in that correspondence, possibly one that depends on µ. This
corresponds to assuming that, when Buyers are indifferent, demand adjusts to meet supply.
I characterize the equilibrium in two steps. I first take as given the laws of motion for beliefs
and find equilibrium prices and quantities and then describe the evolution of beliefs.
5.1
Equilibrium Conditions Given Beliefs
The solution to the entrepreneur’s problem can be found in the same way as in section 4.1.
Entrepreneurs will sell their nonlemons into submarket l, s iff A > pγl,s , so
λM
l,s (pl,s , µl ) =
λl,s (µl )
λl,s (µl ) + (1 − λl,s (µl )) 1 − F pγl,s
(27)
where λl,s (µl ), given by
λµl
λµl + (1 − λ) (1 − µl )
λ (1 − µl )
λl,G (µl ) ≡ Pr [Lemon|l, Green] =
λ (1 − µl ) + (1 − λ) µl
λl,B (µl ) ≡ Pr [Lemon|l, Blue] =
(28)
is the probability that a project is a lemon given that it has emitted signal l, s.
The return obtained by Buyers who purchase from market l, s is
AM
l,s (pl,s , µl ) =
γ
1 − λM
l,s (pl,s , µl )
pl,s
(29)
and it is uncertain for a Buyer because it depends on the true µl .
Lemma 6. Given any beliefs Bl about µl , define λ̂l,s ≡ Pr [Lemon|l, s, Bl ]
1. λ̂l,s = λl,s (µ̂l )
M
2. E[AM
l,s (pl,s , µl |Bl )] = Al,s (pl,s , µ̂l )
The binary structure of both signals and project quality means that the mean of beliefs
about µl is a sufficient statistic for the problem that Buyers care about, which is inferring
project quality from a given signal. This further implies that the expected return from buying
in submarket l, s can be found by simply replacing µ̂l instead of µl in (29). Furthermore, it
means the distribution of µ̂l s across indices l contains all the relevant information about beliefs.
Denote that distribution by H, i.e. let H(m) ≡ Pr[µ̂l ≤ m].
31
Given that there is a continuum of submarkets and beliefs for each submarket are independent, Buyers are able to diversify away all the risk due to realizations of µ (and it will be
optimal for them to do so) and will only care about the expected return when deciding whether
to buy from submarket l, s. The marginal Buyer must be indifferent between buying in any
submarket or investing on his own. Denoting his investment opportunity by A∗ , this implies
that for every l, s
∗
AM
(30)
l,s (pl,s , µ̂l ) ≤ A , with equality if pl,s > 0
∗
If AM
l,s (pl,s , µ̂l ) < A for all pl,s then the l, s market shuts down and the only solution to (30)
is pl,s = 0. This will be the case when λ̂l,s is sufficiently high, either because the signal l, s
is negative or because λ is high and the signal is uninformative. When signals are sufficiently
good (for instance, when s = Green and µ̂l is close to 1), there will be a positive solution to
(30).
General equilibrium is found by equating total spending on buying projects T S (derived
from Buyers’ consumption-savings problem) with total revenue from selling them T R (derived
from the arbitrage condition between selling and keeping projects). Market clearing therefore
requires
ED (p, A∗ ) ≡ T S (p, A∗ ) − T R (p) = 0
(31)
where
Z 1
(1 − β) (1 − λ) γ
T S (p, A ) = K β λ
[µl pl,B + (1 − µl ) pl,G ] dl + (1 − λ) r −
F (A∗ )
∗
A
0
(32)
∗
and
T R (p) = K
Z
1
0

h
i 
γ
pl,B λµl + (1 − λ) (1 − µl ) 1 − F pl,B
+
h
i  dl

γ
pl,G λ (1 − µl ) + (1 − λ) µl 1 − F pl,G
(33)
Given beliefs, conditions (30) and (31) are sufficient to characterize equilibrium prices for any
given state. Given prices, the rest of the equilibrium objects can be derived straightforwardly
from the entrepreneur’s problem. Note, however, that there could be multiple solutions to
equation (30) and the definition of equilibrium does not select between them. Denote the highest
solution to (30) by pl,s (A∗ ) and the vector of such solutions by p(A∗ ). I call an equilibrium
robust if prices satisfy p(X, µ) = p(A∗ ).20
Mild regularity conditions ensure that there exists a unique price vector that satisfies the
conditions for a robust equilibrium.
Lemma 7. If H is continuous, then there is a unique solution to ED(p(A∗ ), A∗ ) = 0.
20
Under Assumption 1, the solution to (30) is always unique. Otherwise, the focus on the highest solution is
justified by the fact the otherwise Buyers could improve their returns by raising prices. See Appendix A for a
discussion of this case.
32
Lemma 7 states that there is a unique equilibrium in which it is never the case that Buyers would prefer higher prices, without requiring Assumption 1 or allowing for rationing in
the definition of equilibrium. Uniqueness is guaranteed because ED(p(A∗ ), A∗ ) is monotonic.
Existence requires the assumption that H be continuous. A small change in A∗ can lead to
a discrete shutdown of a particular submarket, but continuity implies that each submarket is
small so the excess demand function is continuous and must intersect zero at some point. A
similar argument can be found in Riley (1987).
5.2
Evolution of Beliefs
As mentioned above, entrepreneurs can potentially learn about µ from two sources: observing
prices and observing samples from past projects. I first show that there exists an equilibrium where prices do not reveal information and then describe how entrepreneurs learn from
experience.
Lemma 8. There exists a recursive rational expectations equilibrium with signals such that
p(X, µ) does not depend on µ.
µ affects how many lemons and nonlemons end up in each submarket and therefore affects
the supply of projects in each submarket. However, because there is a continuum of submarkets,
demand in each of them is perfectly elastic and these supply realizations have no effect on each
submarket’s price and cancel out in the aggregate. Hence entrepreneurs do not learn about
µ from observing prices and knowing their initial beliefs, summarized by H, is sufficient to
compute equilibrium prices and allocations in any given state.21
The evolution of H is determined by the learning process. After each period, entrepreneurs
use observations of samples from that period to to update beliefs. For each l, these beliefs can
be summarized by a single number, bl,t ≡ Pr [µl,t = µ̄]. µ̂l,t is simply given by
µ̂l,t = bl,t µ̄ + (1 − bl,t )(1 − µ̄)
It is useful to analyze the updating of bl,t in three steps. First, since the number of l-indexed
projects actually sold (and therefore ωl ) depends on the true value of µl , the number of signals itself is a source of information.22 Second, given Nl , each observation can be treated as a
Bernoulli trial, where observing Blue, Lemon or Green, Nonlemon is a success, which happens
with probability µl , and observing Blue, Nonlemon or Green, Lemon is a failure, which happens with probability 1 − µl . Third, entrepreneurs take into account that µl may have changed
21
I leave aside the question of whether there are other equilibria where revealing prices are self-fulfilling.
Quantitatively, this source of information is negligible compared to the information derived from the actual
content of the signals
22
33
between t and t + 1. The solution to this filtering problem can be found analytically. After
observing nl successes out of Nl observations, bl,t+1 is given by:
bl,t+1 =
(1 − σ) µ̄nl (1 − µ̄)Nl −nl (ωlµ̄ )Nl e−ωlµ̄ bl,t + σ (1 − µ̄)nl µ̄Nl −nl (ωl1−µ̄ )Nl e−ωl1−µ̄ (1 − bl,t )
µ̄nl (1 − µ̄)Nl −nl (ωlµ̄ )Nl e−ωlµ̄ bl,t + (1 − µ̄)nl µ̄Nl −nl (ωl1−µ̄ )Nl e−ωl1−µ̄ (1 − bl,t )
where ωlµ̄ and ωl1−µ̄
(34)
denote the values of ωl when µl takes the values µ̄ and 1 − µ̄ respectively.
In order to interpret equation (34), consider the extreme case in which ωl = 0 (which implies
Nl = 0), i.e. entrepreneurs do not observe anything regarding index l. In this case,
bl,t+1 = (1 − σ) bl,t + σ (1 − bl,t )
so bl (and therefore µ̂l ) moves towards 12 , meaning that signals at t + 1 are less informative
than they were at t. The reason for this is that, because there is always a possibility that the
signal structure might change, not learning anything about index l for one period means that
the agents’ understanding of the information structure has become less precise. Experience is
a form of intangible capital, and can depreciate. Conversely, suppose that the realized value
of Nl is very large. The law of large numbers implies that the number of Bernoulli successes
observed will be close to the true µl,t with high probability. In the limit, agents will know
µl,t exactly and bl,t+1 approaches 1 − σ or σ. For intermediate cases, equation (34) implies
that bl will move towards 12 whenever (i) bl,t − 21 is large (mean reversion); (ii) few signals are
observed (experience becomes outdated) or (iii)
with each other).
nl
Nl
is close to
1
2
(different observations conflict
The evolution of µ̂l for each individual l will in general depend on the realizations of the true
µl . However, the following lemma establishes that it is possible to characterize the distribution
of next-period mean beliefs H ′ without knowing the true realized µ.
Lemma 9. H ′ is a deterministic function of X
By Lemma 9, the realized value of µ in any given state is irrelevant not only for the determination of prices and allocations in that state, but also for the learning process. This
makes it possible to characterize the entire dynamic path of the economy by keeping track of
H, aggregate capital K and productivity Z, without any reference to realized µ at all.
Computationally, the only complication is the need to carry the infinite-dimensional state
variable H and compute its trasition density. However, H itself can be well approximated
by a finite grid and its transition density computed by simulation. The fact that prices and
quantities can be found statically means there is no need to compute the entrepreneur’s value
function.
34
5.3
Persistence
It is straightforward to verify that, taking H as given, the comparative statics of the economy
with signals regarding the response to shocks are the same as those in the economy without
signals. In addition, learning introduces a dynamic feedback mechanism between activity in
financial markets and the real economy. Suppose the economy suffers a negative productivity
shock. This lowers r, which lowers demand, increases A∗ and lowers asset prices. At these
lower asset prices, marginal Sellers in each submarket become Keepers, lowering the number
of transactions. Since ωK < ωS , equation (25) implies that the sample sizes from which entrepreneurs will learn about µ will be lower. Equation (34) then implies that this will lead to
a distribution of beliefs H ′ that is more concentrated towards 21 , increasing the overall level
of informational asymmetry as signals become less informative. This will affect asset prices,
the amount of financial market activity, the amount of learning and capital accumulation in
future periods. Thus temporary shocks can have long-lasting real effects. In fact, under certain
conditions a temporary shock can lead to an arbitrarily long recession.
Consider the steady states of two otherwise identical economies, economy 0 with no signals
and economy 1 with signals and endogenous learning. In economy 0, the steady state simply
0
0
consists of a level of capital Kss
such that K ′ = Kss
. In economy 1, the steady state is a level
1
1
1
1
of capital Kss
and a distribution of beliefs Hss
such that K ′ = Kss
and H ′ = Hss
. Denote the
0
1
steady state levels of output in both economies by Yss and Yss . Suppose that the steady state
in economy 0 is such that the market shuts down but in economy 1 there is a positive measure
of submarkets with positive prices. Because economy 0 is like an economy with signals where
µ̂l = 21 for all l, by continuity there exists a minimum number ε0 > 0 such that if a single
submarket had signals with informativeness |µ̂l − 12 | = ε0 the price in that market would be
positive.
Proposition 10.
1. Yss1 > Yss0
2. Fix any integer T > 0 and real number δ > 0. Suppose that, starting from steady state,
economy 1 suffers a negative productivity shock lasting n periods. If
(a) The productivity shock is sufficiently large
(b)
n∈
!
log µ̄ − 12 − log ε0
log Kss − log K0
− 1,
− log (1 − 2σ)
− log [γ (1 − λ)]
(c) ωK is sufficiently small
35
1
0
then there is a T ′ ≥ T such that Yt+T
′ − Yss < δ
Economy 1 will have a higher steady state capital stock and therefore output because
financial markets function, at least partially, allowing high-A entrepreneurs to obtain financing
and low-A entrepreneurs to obtain higher returns, both of which promote capital accumulation.
A large enough negative shock will lead financial markets in economy 1 to shut down. ωK close
to zero means that it is very unlikely that entrepreneurs will observe the outcomes of projects
that were not sold. Therefore a market shutdown will imply an almost complete interruption
of the learning process and entrepreneurs’ understanding of the information structure will
deteriorate. If positive amounts of information are indispensable for trade and the shock lasts
long enough, then when the shock is over financial market activity will not recover because the
information needed to sustain it will have been destroyed. The bounds on n in the statement
of the proposition ensure that the shock lasts long enough for knowledge to depreciate but
not long enough so that the capital stock falls below the informationless steady state level.
Reconstructing the stock of knowledge will require learning mostly from non-sold projects, and
small sample sizes imply that this process will be slow. Hence the levels of output can remain
close to those of the informationless steady state for a long time.
5.4
Simulations
In this section I compute examples of how the economy responds to various shocks, taking into
account the endogenous learning process. The examples are intended as explorations of the
effects that are possible in the model and rough indications of potential magnitudes rather than
as quantitatively precise estimates. To highlight the role of learning, in each case I compare
the impulse responses to those of an economy with no learning where H is fixed at its steady
state value.
Parameter
β
γ
λ
σ
µ̄
F (A)
Y
ωs
ωK
L
Z
Value
0.965
2.14
0.58
0.25
0.74
Gamma distribution with E (A) = 1 and std (A) = 1.75
Z [(1 − λ) K]α L1−α with α = 0.4
200
0.04
1
1
Table 2: Parameter values used in simulations
36
The parameter values I use differ from those in the simulations in section 4.5 because I wish
to focus on economies where information is indispensable for trade, so that markets shut down
when there is no information. I do this by choosing λ = 0.58, which makes lemons abundant
and the asymmetric information problem severe. The length of the period is about a year. The
values of β and γ lead to a steady-state investment-to-GDP ratio of 0.15 and a depreciation rate
of 0.1. σ parameterizes the Markov process followed by µl . Define the half-life of that Markov
process as the number of periods of no learning that it would take for E [µl ] to mean-revert half
log 2
. σ = 0.25 implies a
way back to 21 . A simple calculation shows that it is given by − log(1−2σ)
half-life of 1 period. µ̄ = 0.74 implies that the true correlation between signals an asset quality
is quite high, so learning it well has the potential to greatly reduce informational asymmetry.
ωS and ωK ensure that agents will have many observations to learn from when a submarket is
open and very few when it is closed.
The first simulation is an illustration of Proposition 10. It turns out that in steady state
learning µl,t very precisely provides sufficient information to sustain trade at t+1. Therefore an
l such that the market is open at t will be open at t+1 with very high probability, making market
openness close to an absorbing state. It is not quite absorbing because, due to the assumption
that Nl is Poisson, there is always a small probability there will be few observations and µ̂l will
move towards 21 . Due to the low value of ωK , it is very likely that if a submarket shuts down at
t, nothing will be learnt and it will remain shut at t + 1, making shutdowns nearly absorbing
as well.
Figures 8 and 9 illustrate the response of this economy to a negative 10% productivity shock
taking place at t = 2 and lasting only one period, starting from steady state.
Begin with the evolution of H, shown in figure 9. It is initially highly concentrated at either
µ̂l = (1 − σ) µ̄ + σ (1 − µ̄) = 0.62 (and symmetrically µ̂l = 0.38), where markets are open or
µ̂l = 21 , where markets are shut. The shock is sufficiently large to shut down all markets for
one period; in this period ωl is close to zero for all markets, so µ̂l shifts towards 12 , according
to equation (34). The distribution H becomes more concentrated around 21 , corresponding to
less informative signals. This loss of information turns out to be sufficient to prevent most
markets from reopening at t = 3 when the productivity shock is over. Hence the distribution
H continues to concentrate towards 12 . Eventually the percentage of markets with µ̂l away from
1
2
begins to recover as some observations emerge even from shut markets.
Panel 2 of figure 8 shows the response of output. The response to the initial shock at t = 2
is mechanical and is reverted at t = 3. Then the impact of increased informational asymmetry
on capital accumulation is felt and output drops steadily for several periods. Thus the model
is able to generate the long recessions following financial crises that have been documented
by Cecchetti et al. (2009), Claessens et al. (2009) and Cerra and Saxena (2008). In this
(admittedly extreme) example, output remains close to 5% below its steady state value for over
37
1. TFP
2. Y
−10
Learning
Fixed H
−15
10
20
10
% dev from ss
−5
0
−5
−10
30
t
4. Average A
10
20
0
−10
−20
−30
30
t
5. Measured Solow
10
20
30
t
6. % of projects sold
5
0
−2
−4
10
20
1
0
%
% dev from ss
2
% dev from ss
3. I
5
% dev from ss
% dev from ss
0
−5
−10
30
10
t
20
0.5
0
30
10
t
20
30
t
Figure 8: Transitory TFP shock that leads to market shutdown
H at t=3
40
40
40
0
h(µ)
60
20
20
0.4
0.5
µ
H at t=8
0
0.6
20
0.4
0.5
µ
H at t=16
0
0.6
60
60
40
40
40
20
0
h(µ)
60
h(µ)
h(µ)
H at t=5
60
h(µ)
h(µ)
H at t=1
60
20
0.4
0.5
µ
0.6
0
0.4
0.5
µ
H at t=30
0.6
0.4
0.5
µ
0.6
20
0.4
0.5
µ
0.6
0
Figure 9: Evolution of H after a transitory TFP shock that leads to market shutdown
38
twenty periods.
Increased informational asymmetry affects capital accumulation by lowering both investment, shown on panel 3, and the average rate of transformation of consumption goods into
capital (average A), shown on panel 4. Average A drops because informational asymmetry
interferes with the flow of coconuts for investment towards higher-A entrepreneurs, as both
marginal Buyers and marginal Sellers in each submarket become Keepers. This may provide
an explanation for the pattern identified by Justiniano, Primiceri and Tambalotti (2008b) who
find, in an estimated quantitative model, that productivity in the investment sector is correlated
with disturbances to the functioning of financial markets.
The model is silent about whether entrepreneurs with higher A will transform a given
amount of coconuts into more machines or better machines. In reality it is likely that both
effects are present to some extent. Panel 5 shows the result of the following exercise. Assume
that all the effect of different values of A is due to better machines. An econometrician does not
observe how good the machines are and measures capital formation by just adding investment,
using the steady state average rate of transformation. Using this mismeasured capital stock,
the econometrician then proceeds to compute Solow residuals. Since average A has decreased,
the econometrician’s procedure overestimates capital formation and leads to lower estimated
Solow residuals. This may help explain the long periods of low (measured) productivity growth
that follow some financial crises, as documented for instance by Hayashi and Prescott (2002)
for Japan in the 1990s. In this example, low measured Solow residuals account for about a
quarter of the decrease in output and lower investment accounts for the rest.
The final panel of the figure tracks the drop and then recovery of financial activity as
knowledge of the information structure is first destroyed and then reconstructed.
Beyond the immediate impact, the effects of the productivity shock are due to the deterioration of the economy’s stock of financial knowledge. The next exercise is to consider a shock
to that affects that knowledge directly. Suppose that σ increases from 0.2 to 0.5 for one period.
σ = 0.5 means that Pr [µl,t+1 = µ̄|µl,t = µ̄] = Pr [µl,t+1 = µ̄|µl,t = 1 − µ̄] = 12 . There is a 50%
chance that signals will change meaning between periods, which makes any knowledge of the
time-t information structure irrelevant as of time-t + 1. Effectively, the shock destroys the stock
of financial knowledge. Figures 10 and 11 show that, aside from the initial period, the effects
on the quality of information and therefore on other variables in the economy are very similar
to those of a large productivity shock.
Due to the extreme values of the parameters ωS and ωK , the response of the economy
changes in a highly nonlinear way with the size of the shock. The next exercise (figures 12 and
13) looks at a productivity shock of only 5% rather than 10% which, for these parameter values,
is not enough to shut down the market completely. The number of projects sold decreases in
response to the shock, as seen on panel 6 of figure 12, but is not close to the point where the
39
1. TFP
2. Y
0
Learning
Fixed H
−0.5
−1
10
20
5
% dev from ss
0.5
0
−2
−4
−6
30
t
4. Average A
10
20
0
−5
−10
−15
30
t
5. Measured Solow
10
20
30
t
6. % of projects sold
0.5
0
−1
−2
−3
10
20
1
0
%
% dev from ss
1
% dev from ss
3. I
2
% dev from ss
% dev from ss
1
−0.5
−1
30
10
t
20
0.5
0
30
10
t
20
30
t
Figure 10: A shock that destroys the stock of expertise
H at t=1
H at t=3
40
20
20
0.4
0.5
µ
H at t=8
0
0.6
40
20
0.4
0.5
µ
H at t=16
0
0.6
0.5
µ
0.6
40
0
0.4
0.5
µ
H at t=30
0.6
0.4
0.5
µ
0.6
60
20
0.4
40
20
60
h(µ)
h(µ)
60
0
40
h(µ)
0
60
h(µ)
60
h(µ)
h(µ)
60
H at t=5
40
20
0.4
0.5
µ
0.6
0
Figure 11: Evolution of H after a shock that destroys the stock of expertise
40
market shuts down. Sample sizes for learning about µ decrease roughly in the same proportion
as the drop in the number of sold projects, but since ωS is very high they are still large enough
that there is virtually no effect on the learning process and there is no information-induced
recession.
1. TFP
2. Y
−4
Learning
Fixed H
10
20
0
−2
−4
−6
30
t
4. Average A
20
−5
−10
−15
30
10
20
30
t
6. % of projects sold
2
1
0
−1
10
20
30
1
0
%
% dev from ss
% dev from ss
10
0
t
5. Measured Solow
2
−2
5
% dev from ss
−2
−6
3. I
2
% dev from ss
% dev from ss
0
−2
0.5
−4
−6
10
t
20
t
30
0
10
20
30
t
Figure 12: Transitory TFP shock that does not lead to market shutdown
Learning effects can also lead to high persistence after positive shocks. Consider an economy
with ωK = 0, so there is never any learning from markets that are shut, making market
shutdown an absorbing state.23 Since market openness is not absorbing, the only steady state
of this economy will be one with no trade and no information. However, a positive productivity
shock that led markets to reopen for one period would lead to a large amount of learning,
which could sustain financial market activity for a long time. Figures 14 and 15 illustrate the
response of the economy to such a shock. Thanks to the restarting of the learning process,
information improves a lot at first and, because ωS is high, it depreciates very slowly. This
leads to a sustained increase in output. If the capital stock is computed without adjusting for
the higher average A, around a 20% of the increase in output would be attributed to higher
TFP.
The final experiment consists of modeling a stabilization of the information structure, i.e.
a decrease in σ. This helps the learning process by slowing the rate at which knowledge of the
information becomes outdated. I simulate a permanent decrease in σ from 0.2 to 0.1, using less
23
For this simulation I use λ = 0.53. Markets still shut down in the informationless steady state but the
positive shock that would be required to reopen them is smaller than with λ = 0.58. I set γ = 1.91 so the rate
of depreciation is still 0.1.
41
H at t=1
H at t=3
20
20
0.4
0.5
µ
H at t=8
0
0.6
0.4
0.5
µ
H at t=16
20
0.4
0.5
µ
0
0.6
0.4
0.5
µ
H at t=30
0.6
0.4
0.5
µ
0.6
40
h(µ)
20
0
0
0.6
40
h(µ)
40
20
h(µ)
0
h(µ)
40
h(µ)
40
h(µ)
40
H at t=5
20
0.4
0.5
µ
0
0.6
Figure 13: Evolution of H after a transitory TFP shock that does not lead to market shutdown
1. TFP
2. Y
2
Learning
Fixed H
10
20
10
5
0
30
t
4. Average A
20
30
0
−20
6
5
0
10
20
30
20
30
t
6. % of projects sold
3
4
2
2
1
0
−2
10
%
% dev from ss
% dev from ss
10
20
t
5. Measured Solow
10
−5
40
% dev from ss
4
0
3. I
15
% dev from ss
% dev from ss
6
10
t
20
t
30
0
10
20
t
Figure 14: Positive TFP shock that leads markets to reopen
42
30
H at t=3
40
30
30
30
20
0
0.3
h(µ)
40
10
20
10
0.4
0.5
0.6
µ
H at t=8
0
0.3
0.7
20
10
0.4
0.5
0.6
µ
H at t=16
0
0.3
0.7
40
40
30
30
30
20
10
0
0.3
h(µ)
40
h(µ)
h(µ)
H at t=5
40
h(µ)
h(µ)
H at t=1
20
10
0.4
0.5
µ
0.6
0.7
0
0.3
0.4
0.5
0.6
µ
H at t=30
0.7
20
10
0.4
0.5
µ
0.6
0.7
0
0.3
0.4
0.5
µ
0.6
0.7
Figure 15: Evolution of H after a positive TFP shock that leads markets to reopen
extreme parameters for the Poisson sample sizes: ωS = 3 and ωK = 1. The results are shown
in figures 16 and 17.
As a result of the stabilization, the distribution of H gradually spreads out, improving the
quality of information. This increases the average productivity of investment, leading to a
new steady state with higher output. Around a quarter of the increase in output would be
attributed to higher TFP. This experiment is suggestive of some of the channels by which a
more stable economic environment, which does not need to be re-learned every period can lead
to higher levels of output.
6
Final remarks
This paper has explored the macroeconomic implications of asymmetric information about asset
quality when assets are necessary collateral for financial transactions. Informational asymmetry
acts like a tax on transactions, which has the potential to greatly distort the flow of investment.
Furthermore, the distortions are sensitive to macroeconomic shocks and amplify their effects.
Public information about asset quality may alleviate informational asymmetry, provided
agents have the experience necessary to interpret this information. By modeling the gaining
of experience as the result of financial market activity, the model captures a new notion of
market liquidity that emphasizes an economy’s accumulated financial knowledge. The dynamics of gaining and losing experience can create a powerful propagation mechanism that leads
from temporary shocks to long-lasting consequences for market liquidity, capital-accumulation,
43
1. TFP
2. Y
0
Learning
Fixed H
−0.5
−1
10
20
30
% dev from ss
0.5
5
0
−5
30
t
4. Average A
10
20
20
10
0
−10
30
t
5. Measured Solow
10
20
30
t
6. % of projects sold
3
% dev from ss
10
5
0
−5
10
20
2
2
%
% dev from ss
3. I
10
% dev from ss
% dev from ss
1
1
1
0
−1
30
10
20
t
0
30
10
t
20
30
t
Figure 16: A stabilization of the information structure
H at t=3
2
2
2
0
0.3
h(µ)
3
1
1
0.4
0.5
0.6
µ
H at t=8
0
0.3
0.7
1
0.4
0.5
0.6
µ
H at t=16
0
0.3
0.7
3
3
2
2
2
1
0
0.3
h(µ)
3
h(µ)
h(µ)
H at t=5
3
h(µ)
h(µ)
H at t=1
3
1
0.4
0.5
µ
0.6
0.7
0
0.3
0.4
0.5
0.6
µ
H at t=30
0.7
1
0.4
0.5
µ
0.6
0.7
0
0.3
0.4
0.5
µ
0.6
Figure 17: Evolution of H after a stabilization of the information structure
44
0.7
productivity and output, in ways that are consistent with stylized facts about financial crises.
At the center of the learning mechanism lies an externality: by choosing to sell their projects,
entrepreneurs contribute to the generation of knowledge. The externality is especially strong
when financial markets are close to shutting down. Is there something the government should
do to correct this? The model abstracts from any costs of preparing information for Sellers or
of analyzing it for Buyers. If these costs were literally zero, then it would be simple to compel
agents to produce and analyze information regardless of market conditions, severing the link
between learning and financial activity and eliminating the externality. If instead knowledge
generation is costly and is only undertaken as a side-product of financial transactions, there
may be a case for the government to try to prevent a complete market shutdown in order to
preserve the stock of financial knowledge.
The idea that learning-by-doing about how to interpret information may affect informational
asymmetries could have wider applicability beyond the types of settings explored in this paper.
Exploring whether these mechanisms may account for differing levels of liquidity across different
markets is a promising question for further research.
A
Increasing AM (p) and rationing
Define
pm (p) ≡ arg max AM (p̃)
p̃≥p
p
M
m
≡ p (0)
P m ≡ {p ∈ R+ : p ∈ pm (p)}
pm (p) is the price (or prices) above p that maximize the return for Buyers. If AM (p) has multiple
local maxima, there may be values of p such that pm (p) contains more than one element.
Assumption 2. AM (p) may have many local maxima, but no two are equal
Assumption 2 implies that pm (p) contains at most two elements and if it contains two, one
of them must be p. The results in this appendix would still hold without it, but making this
assumption simplifies the proofs without ruling out any cases of economic interest. pM is the
global maximizer of AM (p), which exists because AM (p) is bounded and continuous and must
be unique by Assumption 2. P m is the set of prices such that Buyers cannot be made better off
with higher prices. Suppose Assumption 1 does not hold, i.e. p∗ ∈
/ P m , where p∗ is defined by
(21). What should one expect from an equilibrium? Stiglitz and Weiss (1981) argue that the
Buyers will offer to pay a price pm (p∗ ) and ration the excess supply (and possibly if AM (p) has
45
multiple maxima, buy the projects rationed out of the market at some lower price pl (p∗ ) < p∗ ).
This is illustrated in Figure 18.
A∗ (p)
b
a
c
pa
pm (pa ) pb
pl (pc )
d
pc
pm (pc )
pd
p
Figure 18: AM (p) and equilibrium prices with rationing
Here P m = [pm (pa ), pl (pc )]∪[pm (pc ), ∞). If the highest Walrasian (nonrationing) equilibrium
price lies at a point like pb ∈ P m or pd ∈ P m , then these are reasonable equilibrium prices.
If the highest Walrasian equilibrium price lies at a point like pa , then Buyers prefer to offer
pm (pa ), which improves the proportion of nonlemons enough to improve their returns. At that
price, there is excess supply, so a fraction of Sellers are rationed out of the market. No matter
how cheaply they offer to sell their projects, no one will be willing to buy them. If the highest
Walrasian equilibrium price lies at a point like pc , then Buyers prefer to raise prices up to
pm (pc ) and ration the excess supply. Unlike case a, if those rationed out of the market offer
to sell their projects at a price below pl (pc ), then this provides a return to Sellers which is
better than that obtained at price pm (pc ). In equilibrium, Buyers anticipate the possibility of
a second round market, which implies that the return from buying in each round must be the
same. Therefore the second-round price must be pl (pc ), such that AM (pm (pc )) = AM (pl (pc )).
The number of projects actually bought in the first round must be exactly such that, given the
projects that remain unsold, the second-round market clears.
Formally, this notion of equilibrium is captured as follows.24 Let ρn (X) be the fraction of
24
Arnold (2005) applies this equilibrium concept to the Stiglitz and Weiss (1981) model
46
Sellers who manage to sell in each of the two rounds, at a price pn (X).The entrepreneur solves
V (k, A, X) =
max
c,k ′ ,l,sL,n ,sNL,n ,dn
[u (c) + βE [V (k ′ , A′ , X ′ ) |X]]
(35)
s.t.
c+l+
X
pn (X) [dn − ρn (X) (sL,n + sN L,n )] ≤ (1 − λ) r (X) k
n=1,2
"
k ′ = γ (1 − λ) k +
X
n=1,2
#
1 − λM
+ Ai
n (X) dn − rn (X) sN L,n
l ≥ 0, dn ≥ 0
sL,1 ∈ [0, λk] , sN L,1 ∈ [0, (1 − λ) k]
sL,2 ∈ [0, λk − ρ1 (X) sL,1 ] , sN L,2 ∈ [0, (1 − λ) k − ρ1 (X) sN L,1 ]
In this formulation, sL,n and sN L,n represent the lemons and nonlemons respectively that the
entrepreneur attempts to sell in round n; he only manages to sell ρn (X) sL,n and ρn (X) sN L,n
respectively.
Supply and demand are defined in the obvious way
SL,n (X) ≡
SN L,n (X) ≡
Z
Z
sL,n (k, A, X) dΓ(k, A)
sN L,n (k, A, X) dΓ(k, A)
Sn (X) ≡ SL,n (X) + SN L,n (X)
Z
Dn (X) ≡
dn (k, A, X) dΓ(k, A)
Definition 3. A recursive competitive equilibrium with rationing consists of prices {pn (X) ,
r (X) , w (X)}; rationing coefficients ρn (X); market proportions of lemons λM (X); a law of
motion Γ′ (X) and associated transition density Π (X ′ |X); a value function V (k, A, X) and
decision rules {cw (X) , c (k, A, X) , k ′ (k, A, X) , l (k, A, X) , sL,n (k, A, X) , sN L,n (k, A, X) ,
dn (k, A, X)} such that (l) factor prices equal marginal products: w (X) = YL (X), r (X) =
YK (X); (ii) workers consume their wage cw (X) = w (X); (iii) {c (k, A, X) , k ′ (k, A, X) ,
l (k, A, X) , sL,n (k, A, X) , sN L,n (k, A, X) , dn (k, A, X)} and V (k, A, X) solve program (35)
′
taking pn (X) , ρn (X) , r (X) , λM
n (X) and Π (X |X) as given; (iv) either (a) the market clears
at a price that Buyers do not wish to increase, i.e. S1 (X) = D1 (X), S2 (X) = D2 (X) = 0,
ρ1 (X) = ρ2 (X) = 1, p1 (X) = p2 (X) ∈ P m (d) there is rationing at pM , i.e. p1 (X) = p2 (X) =
1 (X)
2 (X)
pM , ρ1 (X) = D
≤ 1, ρ2 (X) = D
= 0 or (c) there is rationing in the first round and
S1 (X)
S2 (X)
1 (X)
2 (X)
market clearing in the second, i.e. ρ1 (X) = D
≤ 1, ρ2 (X) = D
= 1, p1 (X) ∈ P m ,
S1 (X)
S2 (X)
p2 (X) ∈ P m; (v) the market proportions of lemons are consistent with individual selling deci47
S
(X)
L,n
sions: λM
n (X) = Sn (X) and (vi) the law of motion of Γ is consistent with individual decisions:
R
Γ′ (k, A)(X) = k′ (k̃,Ã,X)≤k dΓ(k̃, Ã)F (A)
Lemma 10. The equilibrium exists and is unique
Proof. Take any state X and Let A∗ be the investment opportunity of the marginal Buyer.
Total spending on projects is
h
γ i
T S (p1 , p2 , ρ1 , ρ2 , A∗ ) = K β [λ [ρ1 p1 + ρ2 (1 − ρ1 )p2 ] + (1 − λ) r] − (1 − β) (1 − λ) ∗ F (A∗ )
A
and total revenue from sales is

h
i

ρ1 p1 λ + (1 − λ) 1 − F pγ1
+
h
h
ii 
T R (p1 , p2 , ρ1 , ρ2 ) = K 
ρ2 p2 λ + (1 − λ) 1 − F pγ2
− ρ1 λ + (1 − λ) 1 − F pγ1
Equilibrium condition (iv) implies
ED (p1 , p2 , ρ1 , ρ2 , A∗ ) ≡ T S (p1 , p2 , ρ1 , ρ2 , A∗ ) − T R (p1 , p2 , ρ1 , ρ2 ) = 0
The function ED (p1 , p2 , ρ1 , ρ2 , A∗ ) is increasing in A∗ and decreasing in p1 , p2 , ρ1 and ρ2 .
Let
(
the highest solution to AM (p) = A∗ if a solution exists
h
∗
p (A ) ≡
pM
otherwise
(
1 if a solution exists
ρh (A∗ ) ≡
0 otherwise
Both ph (A∗ ) and ρh (A∗ ) are decreasing, which implies that EDh (A∗ ) ≡ ED(ph (A∗ ), ph (A∗ ), ρh (A∗ ),
ρh (A∗ ), A∗ ) is increasing in A∗ . By definition, in equilibrium either EDh (A∗ ) = 0 or EDh (A∗ ) =
0 crosses zero discontinuously at A∗ Since EDh (A∗ ) is increasing, this implies uniqueness.
To establish existence, distinguish three cases:
1. EDh (A∗ ) = 0 for some A∗ . Then the following values constitute an equilibrium: p∗1 =
p∗2 = ph (A∗ ), ρ∗1 = 1, ρ∗2 = 0.
2. EDh (A∗ ) crosses zero discontinuously at A∗ = A∗ (pM ). Then ED pM , pM , 1, 0, A∗ <
0 < ED pM , pM , 0, 0, A∗ so there exists a value of ρ∗1 ∈ (0, 1) such that ED pM , pM , ρ1 , 0, A∗ =
0. Then the following values constitute an equilibrium: p∗1 = p∗2 = pM , ρ∗1 , ρ∗2 = 0
3. EDh (A∗ ) crosses zero discontinuously at some other value of A∗ . This implies that
ph (A∗ ) is discontinuous at A∗ , which, by Assumption 2, implies that AM (p) = A∗
48
must have exactly two solutions in P m , the higher one of which of which is local maximum. Denote them ph (A∗ ) and pl (A∗ ). We have that ED(ph (A∗ ), ph (A∗ ), 1, 1, A∗ ) <
0 < ED pl (A∗ ), pl (A∗ ), 1, 1, A∗ , which implies there is a value of ρ∗1 ∈ (0, 1) such that
ED ph (A∗ ), pl (A∗ ), ρ1 , 1, A∗ = 0. Then the following values constitute an equilibrium:
p∗1 = ph (A∗ ), p∗2 = pl (A∗ ), ρ∗1 , ρ∗2 = 1.
Lemma 11. Consider the equilibrium with signals (Definition ??), given by conditions (31)
and (30) and suppose µl = 12 + ǫ(l − 12 ). In the limit as ǫ → 0, the equilibrium with signals
converges to the rationing equilibrium.
Proof. Suppose in a given state X the rationing equilibrium is given by {A∗ , p1 , p2 , ρ1 , ρ2 }, with
AM (p1 ) = AM (p2 ) = A∗ . Recall that the function AM
l,s (p; µl ) is continuous in µl and equal to
1
M
A (p) when µl = 2 . Consider any δA > 0, δp > 0 and ρ1 , ρ2 ∈ [0, 1]. By continuity, there exists
∗′
∗
ǫ (δA , δp ) small enough that, for any ǫ < ǫ (δA , δp ), there exists A∗′
<
1 (ǫ) satisfying A (ǫ) − A
′
∗
δA such that the fraction of submarkets l, s where the equation AM
l,s (p; µl (ǫ)) = A has a solution
′
′
pl,s (A∗ ) with pl,s (A∗ ) − p1 < δp is exactly ρ1 and the fraction of the remaining submarkets
′
′
′
where the equation AM (p; µl (ǫ)) = A∗ has a solution pl,s (A∗ ) with pl,s (A∗ ) − p2 < δp is
l,s
exactly ρ2 . The result then follows from noting that if for every ǫ a fraction ρ1 of submarkets
have prices p∗l,s (ǫ) satisfying limǫ→0 p∗l,s (ǫ) = p1 , a fraction ρ2 of the remaining ones have prices
′
p∗l,s (ǫ) satisfying limǫ→0 p∗l,s (ǫ) = p2 and the rest have p∗l,s (ǫ) = 0, and limǫ→0 A∗ (ǫ) = A∗ , then
′
lim ED p∗l,s (ǫ), A∗ (ǫ); µl (ǫ) ≡ ED(p1 , p2 , ρ1 , ρ2 , A∗ )
ǫ→0
B
Proofs
Proof of Lemma 1. r(X) does not depend on the distribution of k because Y does not. For
any given p and λM , linearity of the policy functions and the fact that Aj is independent of k j
imply that SL SN L and D do not depend on the distribution of k and therefore neither do the
market clearing values of p(X) and λM (X). Linearity then implies that neither do aggregate
quantities.
Proof of Lemma 3. The first order and envelope conditions are
uc = β max A, AM (X) E [Vk′ (k ′ , A′ , X ′) |X]
Vk (k, A, X) = Wk (k, A, X) uc
49
and the Euler equation is:
uc = β max A, AM (X) E [Wk′ (k ′ , A′ , X ′ ) |X] uc′
With logarithmic preferences, the Euler equation becomes
Wk′ (k ′ , A′ , X ′ )
1
M
= β max A, A (X) E
|X
c
c′
Conjecture that c = aW (k, A, X), which implies
W (k ′ , A′ , X ′ ) = Wk′ (k ′ , A′ , X ′ ) max A, AM (X) (1 − a) W (k, A, X)
and replace in the Euler equation:
1
Wk′ (k ′ , A′ , X ′ )
= β max A, AM (X) E
|X
aW (k, A, X)
aWk′ (k ′ , A′ , X ′ ) max {A, AM (X)} (1 − a) W (k, A, X)
which reduces to a = 1 − β.
Proof of Lemma 4.
Assume there is an entrepreneur for whom the solutions to both programs differ. For Sellers
both programs are identical so it must be that at least one Buyer or Keeper chooses k ′ <
(1 − λ) γk. Then by revealed preference all Buyers choose k ′ < (1 − λ) γk. Replacing in (14)
yields D = 0.
Proof of Proposition 2.
1. This follows immediately from Lemma 4. Whenever the solutions to the two programs
do not coincide, p∗ = 0 satisfies (21), which therefore holds in either case.
2. In the text.
3. Take any X. For sufficiently large p, S(p) > D (p). If there exists a price such that
D(p) ≥ S(p), then the result follows by continuity. If D(p) < S(p) for all p, then p∗ = 0
is a solution.
Proof of Proposition 3.
1. The first order condition for π

if E VW

 1
π=
anything if E VW


0
if E VW
is
(W ′ , A′ , X ′ ) max A, AM Wk (A′ , X ′ ) − R |X > 0
(W ′ , A′ , X ′ ) max A, AM Wk (A′ , X ′ ) − R |X = 0
(W ′ , A′ , X ′ ) max A, AM Wk (A′ , X ′ ) − R |X > 0
50
so Rf must satisfy:
R
f
E VW (W ′ , A′ , X ′ ) max A, AM Wk (A′ , X ′ ) |X
(36)
=
E [VW (W ′ , A′ , X ′ ) |X]
cov VW (W ′ , A′ , X ′ ) , max A, AM Wk (A′ , X ′ ) |X
= E max A, AM Wk (A′ , X ′ ) |X +
E [VW (W ′ , A′ , X ′ ) |X]
when evaluated at π = 1. Using c = (1 − β) W and uc = VW and evaluating at π = 1:
VW (W ′ , A′ , X ′ ) =
=
1
1
=
(1 − β) W ′
(1 − β) [π max {A, AM } Wk (A′ , X ′ ) + (1 − π) Rf ] (W − c)
1
(37)
M
(1 − β) max {A, A } Wk (A′ , X ′ ) (W − c)
Equation (37) implies that the covariance term in (36) is weakly negative, strictly so if
Wk (A′ , X ′ ) is not a constant. Finally, equation (15) implies that Wk (A′ , X ′ ) is indeed
not constant as long as p(X) 6= AMγ(X) ⇐⇒ λM (X) 6= 0.
2. Under symmetrichinformation
nonlemons is pN L =
the price ofi
γ
AM
and the price of lemons
γ
AM (X)
, which does not depend on the realization of
is zero, so Wk = (1 − λ) r (X) +
′
A . If in addition X is deterministic, then Wk is constant and therefore the covariance
term in equation (36) is zero, which gives the result.
Proof of Proposition 4. Take any state X and let r ∗ , p∗ , λM ∗ and AM ∗ represent equilibrium val
ues under asymmetric information in that state. Multiplying supply and demand by 1 − λM ∗
to express them in quantities of nonlemons rather than total projects, market clearing condition
(21) can be reexpressed as
β M∗
γ
∗
∗
M∗
A [λp + (1 − λ) r ] − (1 − β) (1 − λ) F A
K ≤ (1 − λ) 1 − F
γ
p∗
Turn now to the economy with symmetric information and taxes. Virtual wealth is
W (k, A, X) ≡ T + (1 − λ) r (X) + max p (X) ,
γ
max {A, AM (X)}
At price p the supply of projects is S = (1 − λ) 1 − F pγ∗
and tax revenue is
∗
γ
T = τ p (1 − λ) 1 − F
p∗
λM ∗
γ
∗
=
(1 − λ) p 1 − F
M
∗
1−λ
p∗
= λp∗
∗
51
k
γ (1−λM )
γ
The return to buying projects is AM = p∗ (1+τ
=
)
p∗
∗
r = r . Therefore the virtual wealth is
 γ
∗
∗

k
λp
+
(1
−
λ)
r
+
M∗

A
 λp∗ + (1 − λ) r ∗ + Aγ k
W =


 [λp∗ + (1 − λ) (r ∗ + p∗ )] k
= AM ∗ and, because K is the same,
if A ≤ AM ∗
if A ∈ AM ∗ , γp
if A >
γ
p
i
which is the same as with asymmetric information. This implies that demand for nonlemons
is the same as with asymmetric information and the market clearing condition must hold,
confirming that p∗ is an equilibrium price. Since this is true for every state X, programs (13)
and (23) are identical and allocations also coincide.
Proof of Lemma 5.
1. Market clearing implies
dp
=
dτ
∂S
∂τ
∂D
∂p
− ∂D
∂τ
− ∂S
∂p
where
D (p, τ ) =
S (p, τ ) =
AM (p, τ ) =
β M
γ
A (p) τ (1 − λ) p 1 − F
+ (1 − λ) r − (1 − β) (1 − λ) F AM (p)
γ
p
γ
(1 − λ) 1 − F
p
γ
p (1 + τ )
Taking derivatives and substituting:
β
γF
(AM )AM
1+τ
dp
=−
β
AM
dτ
M
γ rF (A ) p
h
i h
h i
i
AM
r − p 1 − F γp
+ βγ AM τ p 1 − F γp
+ r − (1 − β) f AM 1+τ
h
h i
i
h
i <0
M
(AM )τ
+ βγ AM τ p 1 − F γp
+ r − (1 − β) f (AM ) Ap + f γp pγ2 1 − βF1+τ
2. Market clearing implies
dAM
=
dτ
∂S
∂τ
∂D
∂AM
− ∂D
∂τ
∂S
− ∂A
M
where
D AM , τ
S AM , τ
p AM , τ
β M
γ
A
τ (1 − λ) p AM , τ 1 − F
+
(1
−
λ)
r
−
(1
−
β)
(1
−
λ)
F AM
M
γ
p (A , τ )
M
= (1 − λ) 1 − F A (1 + τ )
γ
=
M
A (1 + τ )
=
52
Taking derivatives and substituting:
dAM
=−
dτ
h
i βF (AM )
τ
f AM (1 + τ ) AM 1 − 1+τ
βF AM + (1+τ )2 1 − F AM (1 + τ )
h
h i
i
<0
β
F (AM ) r + β
AM τ p 1 − F γp
+ r − (1 − β) f (AM ) + f (AM (1 + τ )) [1 + τ − τ βF (AM )]
γ
γ
3. Integrating k ′ over all entrepreneurs, K ′ is given by
′
K =
Z
AM
0
+
+
where
Z
γ
p
M
ZA∞
γ
p
M
βA (T + (1 − λ) r) + β (1 − λ) γ dF (A)
[βA (T + (1 − λ) r) + β (1 − λ) γ] dF (A)
[βA (T + (1 − λ) r) + βA (1 − λ) p] dF (A)
γ
T = τ (1 − λ) p 1 − F
p
Taking derivatives:
M


R∞
AM F AM + AM AdF (A) + τ p 1 − F γp
+ r dA
p 1 − F γp
dK ′
dτ
h
i
= (1 − λ) β  hR ∞
R∞
i dp 
dτ
+ γ AdF (A) + τ 1 − F γp + pf γp pγ2 AM F AM + AM AdF (A) dτ
p
Replacing with the expressions from parts 1 and 2 and rearranging:
x
dK ′
(1 − λ) β dτ
=
!
i
i
h
i
R
R
p
τp 1 − F γ
+ r − (1 − β) 1+τ
1−F γ
AM F AM + ∞M AdF (A) − ∞
γ AdF (A)
p
p
A
h
h
i
i
h
i p R
p
β M
− 1+τ
A
τp 1 − F γ
+ r − (1 − β) τ p γ2 AM F AM + ∞M AdF (A) f γ
γ
p
p
A
p
 h
i

R∞
2
γ
M
M
M
M
M
τ
rτ βF A
A F A
+
AdF (A) − A
p 1−F p
τ − 1+τ
βF A
− 1+τ
γ


AMh
+
i f
R
M
p
+ AM F AM + ∞M AdF (A) p 1 − F γ
−
A
r
p




h


+

β M
A
γ
h

 M
f A

A
"
#
β
R
h
i R
F AM
γ
M
M
∞
∞
pr −A
1−F A
1−F γ
+ 1−F γ
AdF (A) −
1− p
1−F γ
γ AdF (A)
p
p
r
p
(1+τ )
AM
p
h i 2
h
i
R
β
τ
+ 1+τ
p 1−F γ
F AM
AM F AM + ∞M AdF (A) − AM
p
γ
A
where
τ
β
β M
γ
γ
M
M
M
x≡ F A
r+ A
τp 1 − F
+ r − (1 − β) f A +f
(1 + τ ) 1 − βF A
>0
γ
γ
p
p
1+τ
Using the market clearing condition and the fact that as τ → 0, F AM → F γp and
f AM → f γp , this expression reduces to
x
dK ′
(1 − λ) β dτ
=
r
r
−β
+ 1 1 − F AM F AM pAM f AM
p
p
Z ∞
2
p
β
−
F AM pr AM 1 − F AM
+ F AM
+ 1 (1 − β)
AdF (A) < 0
γ
r
AM
−
53




Proof of Proposition 5.
1. Fixing p, higher r increases demand but has no effect on supply.
∂[D(p)−S(p)]
< 0 the equilibrium price must rise to restore market clearing. While this
If
∂p
inequality need not hold for every p, it holds at the p that constitutes the highest solution
to (21).
2. The result follows from part 1 and Assumption 1.
3. The result follows from part 1 and (18).
4. By part 1, the terms inside the integrals of equation (22) are increasing in r. By part 3,
AM is decreasing in r. Since both terms inside the integrals are positive but the second
is greater than the first, the results follows.
Proof of Proposition 6. Denote the original equilibrium by p∗ , λM ∗ , AM ∗ and decompose the
effect of an increase in φ into two steps: (i) the effect of increasing φ while decreasing r to
leave φr constant and (ii) the effect
of restoring ro to its original value. For step (i), equation
n p∗ M ∗
M
M
(21) implies that p, λ , A
= φ , λ , φAM ∗ is an equilibrium for any φ. Furthermore,
equation (15) implies that each entrepreneur’s proportional increase in max AM , A is exactly
′
offset by a proportional decrease in virtual wealth and KK does not change with φ. Step (ii)
consists of increasing r, so the results follow from Proposition 5.
Proof of Proposition 7.
1. Rearranging (20):
1
γ (1 − β) (1 − λ)
D (p) =
β [λp + (1 − λ) r] −
F AM (p)
M
p
A (p)
Condition (24) ensures that
D (p) < βλF AM (p)
for any p. Since the supply of lemons from Buyers is λF AM (p) > D (p), there is no
price that equalizes supply and demand, which implies p∗ = 0.
2. First note that AM (p) is bounded because (i) it is continuous in p, (ii) limp→∞ AM (p) = 0
and (iii) using l’Hôpital’s Rule
f
M
lim A (p) = lim
p→0
p→0
54
γ
p
p2
γ 2 (1 − λ)
λ
which must be equal to zero for A to have a finite mean.
Since AM (p) is bounded, condition (24) is met for sufficiently low r, which proves the
result for coconut-productivity shocks. Also because AM (p) is bounded, then
γ
(1
−
λ)
1
−
F
pφ
γ
M
A (p, φ) =
p λ + (1 − λ) 1 − F γ
pφ
converges uniformly to zero as φ → 0, so a sufficiently large project-productivity shock
also ensures that condition (24) is met.
Proof of Proposition 8. For given prices, equation (18) implies that λM ∗ is increasing in λ. In
addition, D (p) − S (p) is decreasing in λ, so p must fall to restore market clearing. By (18),
this reinforces the increase in λM ∗ .
Proof of Proposition 9. The effect of r on each of the endogenous variables in the asymmetric
information economy can be decomposed into the effect it has in the fixed-wedge symmetric
information economy plus the effect of the change in the implicit τ . By part 3 of Proposition
5, the implicit τ is decreasing in r. The inequalities then follow from Lemma 5.
Proof of Lemma 6. Assume w.l.o.g. that s = G and drop the l subscript for clarity.
1.
λ̂G = Pr [Lemon|G]
λ
= Pr [G|Lemon]
Pr [G]
Z
λ
= Pr [G|Lemon, µ] dB(µ) R
Pr [G|µ] dB(µ)
Z
λ
= (1 − µ)dB(µ) R
[λ(1 − µ) + (1 − λ)µ] dB(µ)
λ(1 − µ̂)
=
λ(1 − µ̂) + (1 − λ) µ̂
55
2.
E λM
G (pG , µ) = Pr [Lemon|G, Sold]
=
=
Pr [Sold|Lemon, G] Pr [Lemon|G]
Pr [Sold|G]
λ̂G
λ̂G + 1 − λ̂G 1 − F pγG
Therefore, using part 1,
γ
E AM
1 − E λM
G (pG , µ) =
G (pG , µ)
pG
= AM
G (pG , µ̂)
Proof of Lemma 7. First note that ED(p, A∗ ) is continuous in both arguments. Furthermore,
∗
since AM
l,s (pl,s , µ̂l ) is continuous in µ̂l and H is continuous, then pl,s (A ) can only be discontinuous on a zero-measure set. Therefore ED(p(A∗ ), A∗ ) is continuous.
M
Next, note that, given that AM
l,s (pl,s , µ̂l ) is continuous and limpl,s →∞ Al,s (pl,s ) = 0, it follows
from the definition that pl,s (A∗ ) is decreasing. Therefore, given that ED(p, A∗ ) is increasing in
A∗ and decreasing in pl,s , ED(p(A∗ ), A∗ ) is monotonically increasing.
Finally, note that for sufficiently low A∗ , pl,s (A∗ ) is arbitrarily large for all l, s, so ED(A∗ ) ≡
ED(p(A∗ ), A∗ ) is necessarily negative. For sufficiently high A∗ , pl,s (A∗ ) = 0, so ED(A∗ ) is
necessarily positive. Given that ED(p(A∗ ), A∗ ) is continuous and monotonically increasing, it
must intersect zero exactly once.
Proof of Lemma 8. An equilibrium must satisfy equations (30) and (31). Conjecture that
p(X, µ) does not depend on µ. If so, the beliefs that are required to define µ̂ in equation
(30) must be the beliefs entrepreneurs had at the beginning of the period since prices do not
reveal information. It remains to show that, when prices are defined by p(A∗ ), the value of
A∗ that ensures market-clearing does not depend on realized µ. The market clearing condition
(31) can be rewritten as
∗
∗
ED(p(A ), A , µ) = K
Z
1
0
56
∗
∗
µl ϕl (A )dl + κ(A ) = 0
where
ϕl (A∗ ) ≡βλ (pl,B (A∗ ) − pl,G (A∗ )) F (A∗ )
γ
∗
+pl,B (A ) λ − (1 − λ) 1 − F
pl,B (A∗ )
γ
∗
−pl,G (A ) λ − (1 − λ) 1 − F
pl,G (A∗ )
and
Z 1
γ
∗
κ(A ) = β λ
pl,G (A )dl + (1 − λ)r − (1 − β) (1 − λ) ∗ F (A∗ )
A
0
Z 1
γ
∗
∗
pl,B (A ) (1 − λ) 1 − F
+ λpl,G (A ) dl
+
pl,B (A∗ )
0
∗
Since the values of µl are independent draws from Bl (µl ) and ϕl is bounded, then by the
law of large numbers ED(p(A∗ ), A∗ , µ) is a constant and therefore so is the value of A∗ that
ensures market clearing.
Proof of Lemma 9.
′
H (m) ≡
Z
I [µ̂l,t+1 ≤ m] dl
Z m − (1 − µ̄)
=
I bl,t+1 ≤
dl
2µ̄ − 1
bl,t+1 is a function of the random variables µl , Nl and nl . The distributions of µl , Nl and nl are
a function of the state (up to a reordering of the indices) and the realizations are independent
across l. The result then follows from the law of large numbers.
Proof of Proposition 10.
1. ED(p, A∗ ) is decreasing in p and increasing in A∗ , so for any given level of capital, an
economy where some prices are positive must have higher A∗ than one where all markets
shut down. From the entrepreneur’s problem, k ′ is given by:

Z

k (k, A, X) = k 


′






λ [µl pl,B + (1 − µl ) pl,G ] max{A, A∗ }
+(1 − λ)r max{A, A∗ }
+(1 − λ)µl max {γ, pl,G max{A, A∗ }}
+(1 − λ)(1 − µl ) max {γ, pl,B max{A, A∗ }}
 
 
 
 dl
 
 
which is increasing in both p and A∗ . Hence every entrepreneur in an economy where
some submarkets are open will accumulate more capital than if all submarkets are closed,
57
and therefore the steady state level of capital must be higher.
2. Note first that, by equation (31), excess demand is increasing in r and therefore A∗ is
increasing in K. This means that if ε0 is the minimum value of |µ̂l,n − 21 | needed for there
0
0
to be any trade when K = Kss
, then a value higher than ε0 is needed when K > Kss
.
Since σ > 0, in steady state no signal in the economy is perfectly informative and there is
some residual informational asymmetry in all submarkets. This implies that, by the same
reasoning used in the proof of Lemma 7, a sufficiently large negative productivity shock
lasting n periods will lead all submarkets to shut down for n periods. Suppose ωK = 0,
so there is no learning while markets shut down. Equation (34) implies that n periods
after the shock: for any submarket l
bl,n
1
1
(1 − 2σ)n +
= bl,ss −
2
2
1
1
≤ (1 − 2σ)n+1 +
2
2
This implies that
1
|µ̂l,n − | ≤
2
1
n+1
(1 − 2σ)
[2µ̄ − 1]
2
log(µ̄− 21 )−log ε0
− 1 ensures that |µ̂l,n − 21 | < ε0 for all l and
for all l. Condition n > − log(1−2σ)
Kss −log K0
condition n < log
ensures that Kt+n > K0 . Jointly, this ensures that markets
− log[γ(1−λ)]
do not reopen in period n. If ωK = 0, this means that they will never reopen and Y 1
will converge to Yss0 . For ωK > 0 small enough, the result holds because H ′ is continuous
in ωK and K ′ is continuous in H. proportion of submarkets for all m ≤ T ′ . Since K ′ is
continuous in H, the result follows.
References
Akerlof, G. A.: 1970, The market for “lemons”: Quality uncertainty and the market mechanism,
The Quarterly Journal of Economics 84(3), 488–500.
Arnold, L. G.: 2005, On the possibility of credit rationing in the stiglitz-weiss model. University
of Regensburg Discussion Papers in Economics, No. 403.
Bernanke, B. and Gertler, M.: 1989, Agency costs, net worth, and business fluctuations, American Economic Review 79(1), 14–31.
58
Bernanke, B. S., Gertler, M. and Gilchrist, S.: 1999, The financial accelerator in a quantitative
business cycle framework, in J. B. Taylor and M. Woodford (eds), Handbook of Macroeconomics, Vol. 1 of Handbook of Macroeconomics, Elsevier, chapter 21, pp. 1341–1393.
Bolton, P., Santos, T. and Scheinkman, J. A.: 2009, Outside and inside liquidity, NBER
Working Papers 14867, National Bureau of Economic Research, Inc.
Carlstrom, C. T. and Fuerst, T. S.: 1997, Agency costs, net worth, and business fluctuations:
A computable general equilibrium analysis, The American Economic Review 87(5), 893–910.
Cecchetti, S. G., Kohler, M. and Upper, C.: 2009, Financial crises and economic activity, NBER
Working Papers 15379, National Bureau of Economic Research, Inc.
Cerra, V. and Saxena, S. C.: 2008, Growth dynamics: The myth of economic recovery, The
American Economic Review 98, 439–457(19).
Chari, V. V., Kehoe, P. J. and McGrattan, E. R.: 2007, Business cycle accounting, Econometrica 75(3), 781–836.
Choe, H., Masulis, R. W. and Nanda, V. K.: 1993, Common Stock Offerings Across the Business
Cycle: Theory and Evidence, Journal of Empirical Finance 1(1), 3–31.
Claessens, S., Kose, M. A. and Terrones, M. E.: 2009, What happens during recessions, crunches
and busts?, Economic Policy 24, 653–700.
Daley, B. and Green, B.: 2009, Waiting for news in the dynamic market for lemons. mimeo,
Stanford University.
de Meza, D. and Webb, D. C.: 1987, Too much investment: A problem of asymmetric information, The Quarterly Journal of Economics 102(2), 281–92.
Eisfeldt, A. L.: 2004, Endogenous liquidity in asset markets, Journal of Finance 59(1), 1–30.
Eisfeldt, A. L. and Rampini, A. A.: 2006, Capital reallocation and liquidity, Journal of Monetary Economics 53(3), 369–399.
Fisher, J. D. M.: 2006, The dynamic effects of neutral and investment-specific technology
shocks., Journal of Political Economy 114(3), 413 – 451.
Gorton, G. B. and Metrick, A.: 2009, Securitized banking and the run on repo, NBER Working
Papers 15223, National Bureau of Economic Research, Inc.
Greenwood, J., Hercowitz, Z. and Krusell, P.: 2000, The role of investment-specific technological
change in the business cycle, European Economic Review 44(1), 91 – 115.
59
Hart, O. and Moore, J.: 1994, A theory of debt based on the inalienability of human capital,
The Quarterly Journal of Economics 109(4), 841–879.
Hayashi, F. and Prescott, E. C.: 2002, The 1990s in japan: A lost decade, Review of Economic
Dynamics 5(1), 206 – 235.
Holmström, B. and Tirole, J.: 1998, Private and public supply of liquidity, Journal of Political
Economy 106(1), 1–40.
House, C. L.: 2006, Adverse selection and the financial accelerator, Journal of Monetary Economics 53(6), 1117–1134.
Justiniano, A., Primiceri, G. E. and Tambalotti, A.: 2008a, Investment shocks and business
cycles. FRB of Chicago Working Paper No. 2008-12.
Justiniano, A., Primiceri, G. E. and Tambalotti, A.: 2008b, Investment shocks and the relative
price of investment. Working Paper, Northwestern University.
Kiyotaki, N. and Moore, J.: 1997, Credit cycles, Journal of Political Economy 105(2), 211–48.
Kiyotaki, N. and Moore, J.: 2003, Inside money and liquidity, ESE Discussion Papers 115,
Edinburgh School of Economics, University of Edinburgh.
Kiyotaki, N. and Moore, J.: 2005, 2002 lawrence r. klein lecture liquidity and asset prices,
International Economic Review 46(2), 317–349.
Kiyotaki, N. and Moore, J.: 2008, Liquidity, business cycles and monetary policy. mimeo,
Princeton University.
Kurlat, P.: 2009, The market for lemonade. mimeo, MIT.
Malherbe, F.: 2009, Self-fulfilling liquidity dry-ups. mimeo, ECARES.
Mankiw, N. G.: 1986, The allocation of credit and financial collapse, The Quarterly Journal of
Economics 101(3), 455–70.
Myers, S. C. and Majluf, N. S.: 1984, Corporate financing and investment decisions when firms
have information that investors do not have, Journal of Financial Economics 13(2), 187 –
221.
Ordoñez, G. L.: 2009, Larger crises, slower recoveries: the asymmetric effects of financial
frictions, Technical report.
60
Riley, J. G.: 1987, Credit rationing: A further remark, American Economic Review 77(1), 224–
27.
Stiglitz, J. E. and Weiss, A.: 1981, Credit rationing in markets with imperfect information,
American Economic Review 71(3), 393–410.
van Nieuwerburgh, S. and Veldkamp, L.: 2006, Learning asymmetries in real business cycles,
Journal of Monetary Economics 53(4), 753–772.
Veldkamp, L. L.: 2005, Slow boom, sudden crash, Journal of Economic Theory 124(2), 230–
257.
61